{
    "paper_id": "O14-1002",
    "header": {
        "generated_with": "S2ORC 1.0.0",
        "date_generated": "2023-01-19T08:04:41.825043Z"
    },
    "title": "Investigating Novel Sentence Modeling Techniques for Extractive Speech Summarization",
    "authors": [
        {
            "first": "Shih-Hung",
            "middle": [],
            "last": "\u5289\u58eb\u5f18",
            "suffix": "",
            "affiliation": {},
            "email": ""
        },
        {
            "first": "Kuan-Yu",
            "middle": [],
            "last": "Liu",
            "suffix": "",
            "affiliation": {},
            "email": ""
        },
        {
            "first": "Yu-Lun",
            "middle": [],
            "last": "Chen",
            "suffix": "",
            "affiliation": {},
            "email": "kychen@iis.sinica.edu.tw"
        },
        {
            "first": "Hsin-Min",
            "middle": [],
            "last": "Hsieh",
            "suffix": "",
            "affiliation": {},
            "email": ""
        },
        {
            "first": "Wen-Lian",
            "middle": [],
            "last": "Wang",
            "suffix": "",
            "affiliation": {},
            "email": ""
        },
        {
            "first": "",
            "middle": [],
            "last": "Hsu",
            "suffix": "",
            "affiliation": {},
            "email": "hsu@iis.sinica.edu.tw"
        },
        {
            "first": "Berlin",
            "middle": [],
            "last": "\u9673\u67cf\u7433",
            "suffix": "",
            "affiliation": {},
            "email": "berlin@ntnu.edu.tw"
        },
        {
            "first": "",
            "middle": [],
            "last": "Chen",
            "suffix": "",
            "affiliation": {},
            "email": ""
        }
    ],
    "year": "",
    "venue": null,
    "identifiers": {},
    "abstract": "",
    "pdf_parse": {
        "paper_id": "O14-1002",
        "_pdf_hash": "",
        "abstract": [],
        "body_text": [
            {
                "text": "\u7576\u524d\u4e3b\u6d41\u7684\u7bc0\u9304\u5f0f\u8a9e\u97f3\u6458\u8981\u65b9\u6cd5\u5927\u81f4\u53ef\u5206\u70ba\u4e09\u985e\uff1a(1)\u57fa\u65bc\u6587\u4ef6\u7d50\u69cb\u6216\u8a9e\u53e5\u4f4d\u7f6e\u8cc7 \u8a0a\u4f86\u9078\u53d6\u91cd\u8981\u8a9e\u53e5\uff1b(2)\u57fa\u65bc\u7279\u5b9a\u8a5e\u5f59\u6216\u8a9e\u610f\u8cc7\u8a0a\u4e4b\u975e\u76e3\u7763\u5f0f(Unsupervised)\u6458\u8981\u65b9\u6cd5\uff1b (3)\u9700\u8981\u4f7f\u7528\u4eba\u5de5\u6458\u8981\u6a19\u8a3b\u4f86\u8a13\u7df4\u6a21\u578b\u4e4b\u76e3\u7763\u5f0f(Supervised)\u6458\u8981\u65b9\u6cd5\u3002\u7b2c\u4e00\u985e\u6458\u8981\u65b9\u6cd5 \u5927\u90fd\u662f\u7c21\u55ae\u5730\u5f9e\u6587\u4ef6\u7684\u7dd2\u8ad6(Introductoion)\u6216\u7d50\u8ad6(Conclusion)\u6240\u5728\u6bb5\u843d\u64f7\u53d6\u51fa\u82e5\u5e72\u8a9e\u53e5 \u4f86\u7d44\u6210\u6458\u8981 [1] \uff1b\u6b64\u985e\u65b9\u6cd5\u50c5\u9069\u7528\u5728\u7279\u5b9a\u5177\u6709\u4e00\u81f4\u7d50\u69cb\u7684\u6587\u5b57\u6216\u8a9e\u97f3\u6587\u4ef6\u4e0a\uff0c\u56e0\u6b64\u5728\u5be6 \u969b\u61c9\u7528\u4e0a\u6709\u5176\u4fb7\u9650\u6027\u3002\u53e6\u4e00\u65b9\u9762\uff0c\u7b2c\u4e8c\u985e\u6458\u8981\u65b9\u6cd5\u901a\u5e38\u5c07\u81ea\u52d5\u6458\u8981\u4efb\u52d9\u8996\u70ba\u5982\u4f55\u6392\u5e8f\u4e26 \u6311\u9078\u5177\u4ee3\u8868\u6027(\u6216\u91cd\u8981\u6027)\u8a9e\u53e5\u4e4b\u554f\u984c\uff0c\u5176\u65b9\u6cd5\u901a\u5e38\u662f\u4ee5\u975e\u76e3\u7763\u5f0f\u65b9\u5f0f\u7522\u751f\u51fa\u4e00\u7a2e\u8a9e\u53e5\u5c64 \u6b21\u7684\u6458\u8981\u7279\u5fb5\u6216\u91cd\u8981\u5206\u6578\u4ee5\u4f9b\u8a9e\u53e5\u6392\u5e8f\u4f7f\u7528\u3002\u800c\u7b2c\u4e8c\u985e\u6458\u8981\u65b9\u6cd5\u53c8\u53ef\u9032\u4e00\u6b65\u5730\u5206\u6210\u4e09\u5c0f \u985e \uff1a (I) \u4ee5 \u5411 \u91cf (Vector) \u70ba \u57fa \u790e \u7684 \u65b9 \u6cd5 \uff0c \u5305 \u542b \u6709 \u5411 \u91cf \u7a7a \u9593 \u6a21 \u578b (Vector Space Model, VSM) [8] ",
                "cite_spans": [
                    {
                        "start": 193,
                        "end": 196,
                        "text": "[1]",
                        "ref_id": "BIBREF0"
                    },
                    {
                        "start": 428,
                        "end": 431,
                        "text": "[8]",
                        "ref_id": "BIBREF7"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            },
            {
                "text": "EQUATION",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [
                    {
                        "start": 0,
                        "end": 8,
                        "text": "EQUATION",
                        "ref_id": "EQREF",
                        "raw_str": "\u7576\u7d66\u5b9a\u4e00\u7bc7\u88ab\u6458\u8981\u6587\u4ef6 D \u6642\uff0c\u6587\u4ef6\u4e2d\u6bcf\u4e00\u53e5\u8a9e\u53e5 S \u7684\u4e8b\u5f8c\u6a5f\uf961 ) | ( D S P \u53ef\u4ee5\u7528\u4f86\u8868 \u793a\u8a9e\u53e5 S \u5c0d\u65bc\u6587\u4ef6 D \u7684\u91cd\u8981\u6027\u3002\u7576\u4f7f\u7528\u8a9e\u8a00\u6a21\u578b\u4f86\u8a08\u7b97 ) | ( D S P \u6642\uff0c\u6211\u5011\u900f\u904e\u8c9d\u6c0f\u5b9a\u7406 (Bayes' Theorem)\u5c07 ) | ( D S P \u5c55\u958b\u6210[3]\uff1a , ) ( ) ( ) | ( ) | ( D P S P S D P D S P \uf03d (1) \u5176\u4e2d ) (D P \u662f\u6587\u4ef6 D \u7684\u4e8b\u524d\u6a5f\uf961\uff0c\u7531\u65bc ) (D P \u4e0d\u5f71\u97ff\u8a9e\u53e5\u7684\u6392\u5e8f\u7d50\u679c\uff0c\u6545\u53ef\u7701\u7565\u4e0d\u8a0e\u8ad6\uff1b \u53e6\u4e00\u65b9\u9762\uff0c ) (S P \u662f\u8a9e\u53e5 S \u7684\u4e8b\u524d\u6a5f\u7387\uff0c\u53ef\u4ee5\u4f7f\u7528\u5404\u5f0f\u975e\u76e3\u7763\u5f0f\u65b9\u6cd5\u6216\u76e3\u7763\u5f0f\u65b9\u6cd5\u4f86\u6c42 \u5f97[3]\u3002\u5728\u6b64\u5148\u5047\u8a2d\u8a9e\u53e5\u7684\u4e8b\u524d\u6a5f\uf961\u70ba\u4e00\u500b\u5747\u52fb\u5206\u5e03(Uniform Distribution)\uff0c\u6240\u4ee5 ) (S P \u4ea6 \u53ef\u7701\u7565\u3002\u6700\u5f8c\uff0c ) | ( S D P \u662f\u8a9e\u53e5 S \u6240\u5f62\u6210\u7684\u8a9e\u8a00\u6a21\u578b\u751f\u6210\u6587\u4ef6 D \u4e4b\u6a5f\uf961(\u6216\u7a31\u4f5c\u6587\u4ef6\u76f8\u4f3c \u5ea6)\uff0c\u53ef\u4ee5\u7528\u4f86\u8868\u793a\u6587\u4ef6 D \u8207\u8a9e\u53e5 S \u4e4b\u9593\u7684\u76f8\u4f3c\u95dc\u4fc2\uff0c\u5982\u679c\u8a9e\u53e5 S \u751f\u6210\u6587\u4ef6 D \u7684\u6a5f\u7387\u503c \u6108\u9ad8\uff0c\u4ee3\u8868\u8a9e\u53e5 S \u8207\u6587\u4ef6 D \u6108\u70ba\u76f8\u4f3c(\u8a9e\u53e5\u6108\u80fd\u4ee3\u8868\u6587\u4ef6 D )\uff0c\u5373\u6108\u6709\u53ef\u80fd\u662f\u6458\u8981\u8a9e\u53e5\u3002 \u6211\u5011\u53ef\u4ee5\u66f4\u9032\u4e00\u6b65\u5730\u5047\u8a2d\u6587\u4ef6 D \u4e2d\u8a5e\u5f59\u8207\u8a5e\u5f59\u4e4b\u9593\u662f\u7368\uf9f7\u7684\uff0c\u4e26\u4e14\u4e0d\u8003\u616e\u6bcf\u4e00\u500b\u8a5e\u5f59\u5728 \u6587\u4ef6 D \u4e2d\u767c\u751f\u7684\u9806\u5e8f\u95dc\u4fc2(\u5373\u8a5e\u888b\u5047\u8a2d(Bag-of-Word Assumption))\uff0c\u5247\u8a9e\u53e5 S \u751f\u6210\u6587\u4ef6 D \u7684\u6587\u4ef6\u76f8\u4f3c\u5ea6\u91cf\u503c(Document Likelihood Measure, DLM) ) | ( S D P \u53ef\u62c6\u89e3\u6210\u6587\u4ef6 D \u4e2d\u6bcf\u4e00 \u7684\u8a5e\u5f59 w \u500b\u5225\u767c\u751f\u7684\u689d\u4ef6\u6a5f\uf961\u4e4b\u9023\u4e58\u7a4d\uff1a , ) | ( ) | ( ) , ( \uf0d5 \uf0ce \uf03d D w D w C S w P S D P (2) \u6b64\u7a2e\u65b9\u6cd5\u662f\u70ba\u8a9e\u53e5 S \u5efa\uf9f7\u4e00\u500b\u8a9e\u53e5\u6a21\u578b(Sentence Model) ) | ( S w P \uff0c w \u662f\u4e00\u500b\u51fa\u73fe\u5728\u6587\u4ef6 D \u4e2d\u7684\u8a5e\u5f59\uff0c ) , ( D w C \u662f\u8a5e\u5f59 w \u51fa\u73fe\u5728\u6587\u4ef6 D \u4e2d\u7684\u6b21\u6578\u3002\u5176\u4e2d\uff0c\u6211\u5011\u53ef\u5229\u7528\u6700\u5927\u5316\u76f8\u4f3c \u5ea6\u4f30\u6e2c(Maximum Likelihood Estimation, MLE)\u7684\u65b9\u5f0f\u4f86\u5efa\u7acb\u6bcf\u4e00\u500b\u8a9e\u53e5\u7684\u8a9e\u53e5\u6a21\u578b\uff1a , | | ) , ( ) | ( S S w C S w P \uf03d (3) \u5728(3)\u4e2d\uff0c ) , ( S w C \u8868\u793a\u8a5e\u5f59 w \u5728\u8a9e\u53e5 S \u4e2d\u51fa\u73fe\u7684\u6b21\u6578\uff0c S \u5247\u8868\u793a\u8a9e\u53e5 S \u6240\u542b\u7684\u7e3d\u8a5e\u6578\u3002 \u503c\u5f97\u6ce8\u610f\u7684\u662f\uff0c\u7531\u65bc\u8a9e\u53e5 S \u901a\u5e38\u50c5\u7531\u5c11\u6578\u5b57\u8a5e\u6240\u7d44\u6210\uff0c\u56e0\u6b64\u5bb9\u6613\u906d\u9047\u8cc7\u6599\u7a00\u758f(Data Sparseness)\u7684\u554f\u984c\uff0c\u9019\u6703\u4f7f\u5f97\u8a9e\u53e5\u6a21\u578b\u4f7f\u7528\u6700\u5927\u5316\u76f8\u4f3c\u5ea6\u4f30\u6e2c\u6642\uff0c\u4e0d\u50c5\u53ef\u80fd\u7121\u6cd5\u6e96\u78ba\u5730 \u4f30\u6e2c\u6bcf\u4e00\u500b\u8a5e\u5f59\u5728\u8a9e\u53e5\u4e2d\u771f\u6b63\u7684\u6a5f\uf961\u5206\u4f48\uff0c\u4e5f\u53ef\u80fd\u56e0\u70ba\u67d0\u4e9b\u8a5e\u5f59\u7684\u689d\u4ef6\u6a5f\u7387\u503c\u70ba\u96f6\uff0c\u5c0e \u81f4\u8a9e\u53e5 S \u7522\u751f\u6587\u4ef6 D \u7684\u6a5f\uf961\u503c\u70ba\u96f6\u3002\u70ba\u4e86\u6e1b\u8f15\u4e0a\u8ff0\u7684\u73fe\u8c61\uff0c\u672c\uf941\u6587\u4f7f\u7528 Jelinek-Mercer \u5e73 \u6ed1 \u5316 (Smoothing) \u6280 \u8853 \u85c9 \u7531 \u4f7f \u7528 \u4ee5 \u5927 \u91cf \u6587 \u5b57 \u8a9e \u6599 \u8a13 \u7df4 \u800c \u6210 \u7684 \u80cc \u666f \u55ae \u9023 \u8a9e \u8a00 \u6a21 \u578b (Background Unigram Language Model)\u4f86\u8abf\u9069\u8a9e\u53e5\u6a21\u578b[33]\uff0c\u6545 ) | ( S D P \u53ef\u9032\u4e00\u6b65\u5730\u8868\u793a \u6210\uff1a , )] | ( ) 1 ( ) | ( [ ) | ( ) , ( \uf0d5 \uf0ce \uf0d7 \uf02d \uf02b \uf0d7 \uf03d D w D w C BG w P S w P S D P \uf06c \uf06c (4) \u5176\u4e2d\uff0c P(w|BG)\u662f\u8a5e\u5f59 w \u7531\u80cc\u666f\u55ae\u9023\u8a9e\u8a00\u6a21\u578b\u6240\u7522\u751f\u7684\u6a5f\uf961\u503c\u3002 2.2\u3001\u5eab\u723e\u8c9d\u514b-\u840a\u4f2f\u52d2\u96e2\u6563\u5ea6\u91cf\u503c \u8a9e\u8a00\u6a21\u578b\u4f7f\u7528\u65bc\u6587\u4ef6\u6458\u8981\u7684\u7814\u7a76\u4e2d\uff0c\u9664\u4e86\u53ef\u88ab\u7528\u65bc\u8a08\u7b97\u8a9e\u53e5\u751f\u6210\u6587\u4ef6\u7684\u53ef\u80fd\u6027\u5916\uff0c\u53e6\u4e00 \u7a2e\u65b9\u5f0f\u70ba\u85c9\u7531\u5eab\u723e\u8c9d\u514b-\u840a\u4f2f\u52d2\u96e2\u6563\u5ea6\u91cf\u503c(Kullback-Leibler Divergence Measure, KL)\uff0c \u4f86\u8a55\u4f30\u6587\u4ef6\u4e2d\u6bcf\u4e00\u500b\u8a9e\u53e5\u7684\u91cd\u8981\u6027\u3002\u7576\u4f7f\u7528\u5eab\u723e\u8c9d\u514b-\u840a\u4f2f\u52d2\u96e2\u6563\u5ea6\u91cf\u503c\u65bc\u6458\u8981\u4efb\u52d9 \u4e2d\uff0c\u88ab\u6458\u8981\u6587\u4ef6 D \u548c D \u4e2d\u7684\u6bcf\u4e00\u500b\u8a9e\u53e5 S \u90fd\u5c07\u5206\u5225\u88ab\u63cf\u8ff0\u70ba\u4e00\u500b\u55ae\u9023\u8a9e\u8a00\u6a21\u578b\uff1b\u7576\u76f8\u5c0d \u65bc\u88ab\u6458\u8981\u6587\u4ef6 D \u7684\u6587\u4ef6\u6a21\u578b(Document Model)\uff0c\u8a9e\u53e5\u6a21\u578b\u7684\u96e2\u6563\u5ea6\u91cf\u503c\u6108\u5c0f\u6642\uff0c\u5247\u4ee3\u8868 \u8a9e\u53e5\u8207\u6587\u4ef6\u6108\u76f8\u95dc\uff0c\u4ea6\u5373\u8a9e\u53e5 S \u6108\u91cd\u8981\u3002\u5728\u6b64\u6458\u8981\u67b6\u69cb\u4e0b\uff0c\u6392\u5e8f\u8a9e\u53e5\u91cd\u8981\u6027\u7684\u516c\u5f0f\u5982\u4e0b [15]\uff1a , ) | ( ) | ( log ) | ( ) || ( \uf0e5 \uf0ce \uf03d V w S w P D w P D w P S D KL",
                        "eq_num": "(5"
                    }
                ],
                "section": "",
                "sec_num": null
            },
            {
                "text": "EQUATION",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [
                    {
                        "start": 0,
                        "end": 8,
                        "text": "EQUATION",
                        "ref_id": "EQREF",
                        "raw_str": "\u9593\u9694\u5f0f\u6700\u9ad8 K \u9078\u53d6\u6cd5\u5c31\u662f\u5728\u6700\u9ad8\u6392\u5e8f\u6587\u4ef6\u4e2d\u6bcf\u9593\u9694 J(\u4f8b\u5982\u5169\u500b\u9593\u9694\uff0cJ \u70ba 2)\u6311\u9078\u51fa K \u500b\u76f8\u95dc\u6587\u4ef6\u51fa\u4f86\u7576\u4f5c\u662f\u6700\u9ad8\u6392\u5e8f\u6587\u4ef6\uff0c\u7c21\u55ae\u7684\u4f8b\u5b50\u5982\u4e0b\uff1a\u5047\u8a2d\u6700\u9ad8\u6392\u5e8f\u6587\u4ef6\u6709 10 \u7bc7\uff0c \u6211\u5011\u6bcf\u9593\u9694 2 \u7bc7\u8981\u6311\u51fa\u6700\u9ad8 3 \u7bc7\u51fa\u4f86\uff0c\u5247\u7b2c\u4e00\u3001\u7b2c\u56db\u53ca\u7b2c\u4e03\u7bc7\u6703\u88ab\u6311\u9078\u51fa\u4f86\u7576\u4f5c\u662f\u6700\u9ad8 \u6392\u5e8f\u6587\u4ef6\u3002\u9593\u9694\u5f0f\u6700\u9ad8 K \u9078\u53d6\u6cd5\u7684\u4e3b\u8981\u601d\u60f3\u662f\u8981\u6311\u9078\u51fa\u5177\u6709\u591a\u5143\u6027\u7684\u6587\u4ef6\uff0c\u4f46\u5176\u5be6\u6b64 \u9078\u53d6\u65b9\u6cd5\u4e5f\u662f\u76f8\u7576\u4e0d\u7a69\u5b9a\u7684\u3002\u7fa4\u4e2d\u5fc3\u9078\u53d6\u6cd5\u5247\u662f\u5148\u5c07\u6700\u9ad8\u6392\u5e8f\u6587\u4ef6\u4f5c\u5206\u7fa4(Clustering)\uff0c \u5206\u7fa4\u65b9\u6cd5\u53ef\u4ee5\u662f\u4efb\u610f\u7684\uff0c\u5e38\u7528\u7684\u5206\u7fa4\u65b9\u6cd5\u70ba K \u4e2d\u5fc3(K-means)\u5206\u7fa4\u6cd5\uff0c\u7136\u5f8c\u518d\u5f9e\u5206\u51fa\u4f86 \u7684\u6bcf\u4e00\u7fa4\u4e2d\u6311\u9078\u51fa\u4e00\u7bc7\u6700\u76f8\u95dc\u7684\u6587\u4ef6\uff0c\u4ee5\u6b64\u69cb\u6210\u65b0\u7684\u6700\u9ad8\u6392\u5e8f\u6587\u4ef6\uff0c\u7531\u5206\u7fa4\u7684\u89c0\u5ff5\u53ef \u77e5\uff0c\u7fa4\u4e2d\u5fc3\u9078\u53d6\u6cd5\u65e8\u5728\u9078\u53d6\u51fa\u5177\u6709\u591a\u5143\u6027\u7684\u6587\u4ef6\uff0c\u8207\u9593\u9694\u5f0f\u6700\u9ad8 K \u9078\u53d6\u6cd5\u76f8\u8f03\uff0c\u7fa4\u4e2d \u5fc3\u9078\u53d6\u6cd5\u662f\u4e00\u500b\u6bd4\u8f03\u7a69\u5b9a\u7684\u9078\u53d6\u65b9\u6cd5\u3002\u4e3b\u52d5\u5f0f-\u95dc\u806f\u591a\u5143\u5bc6\u5ea6\u9078\u53d6\u6cd5\u70ba\u540c\u6642\u8003\u91cf\u6700\u9ad8\u6392 \u5e8f\u6587\u4ef6\u4e2d\u7684\u95dc\u806f\u6027\u3001\u591a\u5143\u6027\u4ee5\u53ca\u5bc6\u5ea6\u6027\u7684\u4e00\u7a2e\u8caa\u5a6a(Greedy)\u9078\u53d6\u6cd5[31]\u3002\u4ee5\u4e0a\u9078\u53d6\u65b9\u6cd5 \u5e38\u898b\u65bc\u8cc7\u8a0a\u6aa2\u7d22\u9818\u57df\u4e2d\uff0c\u6709\u8208\u8da3\u7684\u8b80\u8005\u53ef\u53c3\u8003\u76f8\u95dc\u6587\u737b[5]\uff0c\u672c\u8ad6\u6587\u662f\u9996\u6b21\u5c07\u4e0a\u8ff0\u65b9\u6cd5 \u904b\u7528\u5728(\u8a9e\u97f3)\u6587\u4ef6\u6458\u8981\u4efb\u52d9\u4e2d\u4e26\u505a\u6df1\u5165\u63a2\u8a0e\u3002 \u4e09\u3001\u65b0\u7a4e\u5f0f\u865b\u64ec\u76f8\u95dc\u6587\u4ef6\u9078\u53d6 \u57fa\u65bc\u4e3b\u52d5\u5f0f-\u95dc\u806f\u591a\u5143\u5bc6\u5ea6\u9078\u53d6\u6cd5\uff0c\u9664\u4e86\u8003\u91cf\u5230\u865b\u64ec\u76f8\u95dc\u6587\u4ef6(\u6700\u9ad8\u6392\u5e8f\u6587\u4ef6)\u4e2d\u7684\u95dc\u806f \u6027\u3001\u591a\u5143\u6027\u4ee5\u53ca\u5bc6\u5ea6\u6027\u4e4b\u5916\uff0c\u6211\u5011\u8a8d\u70ba\u975e\u76f8\u95dc\u6027(Non-relevance)\u8cc7\u8a0a(\u5728\u9019\u88e1\u662f\u6307\u8a9e\u53e5\u7684 \u975e\u76f8\u95dc\u6027\u8cc7\u8a0a)\u4e5f\u662f\u76f8\u7576\u91cd\u8981\u7684\u7dda\u7d22\uff0c\u53ef\u4ee5\u7528\u4f86\u5e6b\u52a9\u91cd\u65b0\u9078\u53d6\u66f4\u597d\u7684\u865b\u64ec\u76f8\u95dc\u6587\u4ef6\uff0c\u56e0 \u6b64\u672c\u8ad6\u6587\u63d0\u51fa\u984d\u5916\u8003\u91cf\u975e\u76f8\u95dc\u8cc7\u8a0a\u4ee5\u6539\u9032\u4e3b\u52d5\u5f0f-\u95dc\u806f\u591a\u5143\u5bc6\u5ea6\u9078\u53d6\u6cd5\uff0c\u7a31\u4e4b\u70ba\u4e3b\u52d5\u5f0f- \u95dc\u806f\u591a\u5143\u5bc6\u5ea6\u975e\u76f8\u95dc(Active-RDDN)\u9078\u53d6\u6cd5\u3002\u53e6\u4e00\u65b9\u9762\uff0c\u672c\u8ad6\u6587\u4e5f\u63d0\u51fa\u4f7f\u7528\u91cd\u758a\u5206\u7fa4 (Overlapped Cluster)\u7684\u6982\u5ff5\u4f86\u5e6b\u52a9\u91cd\u65b0\u9078\u53d6\u66f4\u6709\u6548\u7684\u865b\u64ec\u76f8\u95dc\u6587\u4ef6\uff0c\u8332\u4ecb\u7d39\u5982\u4e0b\uff1a 3.1\u3001\u4e3b\u52d5\u5f0f-\u95dc\u806f\u591a\u5143\u5bc6\u5ea6\u975e\u76f8\u95dc\u9078\u53d6 \u5047\u8a2d\u8a9e\u53e5 S \u5df2\u7d93\u8f38\u5165\u5230\u8cc7\u8a0a\u6aa2\u7d22\u7cfb\u7d71\u4e2d\u4e26\u5f97\u5230\u4e86\u6700\u9ad8\u6392\u5e8f\u6587\u4ef6 D Top ={D 1 ,D 2 ,\u2026,D M }\uff0c\u90a3 \u4e3b\u52d5\u5f0f-\u95dc\u806f\u591a\u5143\u5bc6\u5ea6\u975e\u76f8\u95dc\u9078\u53d6\u5247\u662f\u5f9e\u6700\u9ad8\u6392\u5e8f\u6587\u4ef6\u4e2d\u8fed\u4ee3\u5730(Iteratively)\u540c\u6642\u8003\u616e\u56db \u7a2e\u91cd\u8981\u56e0\u7d20(\u95dc\u806f\u6027\u3001\u591a\u5143\u6027\u3001\u5bc6\u5ea6\u6027\u4ee5\u53ca\u975e\u76f8\u95dc\u8cc7\u8a0a)\u4f86\u91cd\u65b0\u9078\u53d6\u66f4\u5177\u4ee3\u8868\u6027\u7684\u6587\u4ef6 \u96c6\u3002\u5177\u9ad4\u5730\u8aaa\uff0c\u6700\u9ad8\u6392\u5e8f\u6587\u4ef6\u4e2d\u7684\u6bcf\u500b\u5019\u9078(Candidate)\u6587\u4ef6 D m \u90fd\u6703\u6709\u8457\u540c\u6642\u8003\u91cf\u56db\u7a2e \u56e0\u7d20\u7684\u4e00\u500b\u7dda\u6027\u7d50\u5408\u7684\u5206\u6578\uff0c\u5176\u9078\u53d6\u516c\u5f0f\u5982\u4e0b\uff1a \uf028 \uf029 \uf028 \uf029 \uf05b \uf028 \uf029 \uf028 \uf029 \uf028 \uf029\uf05d, , , 1 max arg P Top * m Density m Diversity m NR m Rel D D M D M D S M D S M D m \uf0d7 \uf02b \uf0d7 \uf02b \uf0d7 \uf02b \uf0d7 \uf02d \uf02d \uf02d \uf03d \uf02d \uf0ce \uf067 \uf062 \uf061 \uf067 \uf062 \uf061 D D (6) \u5176\u4e2d D P \u70ba\u5df2\u7d93\u9078\u5165\u7684\u6587\u4ef6\u96c6\uff0cM Rel (S,D m )\u3001M NR (S,D m )\u3001M Diversity (D m )\u53ca M Density (D m )\u5206\u5225 \u4ee3\u8868\u5019\u9078\u6587\u4ef6 D m \u7684\u95dc\u806f\u6027\u91cf\u503c\u3001\u975e\u76f8\u95dc\u8cc7\u8a0a\u6027\u91cf\u503c\u3001\u591a\u5143\u6027\u91cf\u503c\u3001\u4ee5\u53ca\u5bc6\u5ea6\u6027\u91cf\u503c\uff0c \u800c \u03b1\u3001\u03b2\u3001\u03b3 \u70ba\u53ef\u8abf\u53c3\u6578\u4e14\u5176\u7e3d\u548c\u70ba 1(\u5373 \u03b1+\u03b2+\u03b3=1)\u3002\u503c\u5f97\u4e00\u63d0\u7684\u662f\uff0c\u5f0f(6)\u8207\u65e9\u671f\u7528\u65bc\u8cc7 \u8a0a\u6aa2\u7d22\u53ca\u6458\u8981\u9818\u57df\u7684\u7d93\u5178\u516c\u5f0f\u6700\u5927\u908a\u969b\u95dc\u806f(Maximal Marginal Relevance, MMR)\u76f8\u4f3c [2]\u3002\u53e6\u5916\uff0c\u95dc\u806f\u6027\u91cf\u503c M Rel (S, D m )\u53ef\u5b9a\u7fa9\u70ba\u6587\u4ef6 D m \u8a9e\u53e5 S \u7684\u8ca0\u5eab\u723e\u8c9d\u514b-\u840a\u4f2f\u52d2\u96e2\u6563\u5ea6 \u91cf\u503c(\u5373-KL(D m ||S))\u3002\u4e0b\u9762\u5c07\u5206\u5225\u4ecb\u7d39\u975e\u76f8\u95dc\u8cc7\u8a0a\u91cf\u503c\u3001\u591a\u5143\u6027\u91cf\u503c\u3001\u4ee5\u53ca\u5bc6\u5ea6\u6027\u91cf\u503c\u3002 3.1.1\u3001\u975e\u76f8\u95dc\u6027\u8cc7\u8a0a\u91cf\u503c \u5c0d\u65bc\u4e00\u500b\u8a9e\u53e5 S\uff0c\u5176\u975e\u76f8\u95dc\u6027(Non-relevance)\u8cc7\u8a0a\u901a\u5e38\u53ef\u4ee5\u5f9e\u7b2c\u4e00\u6b21\u8cc7\u8a0a\u6aa2\u7d22\u6642\u6392\u5728\u6700 \u5f8c\u9762\u7684\u4e00\u4e9b\u6587\u4ef6(\u6700\u4f4e\u6392\u5e8f(Low-Ranked)\u6587\u4ef6)\u4f86\u4ee3\u8868\uff0c\u90a3\u9ebc\u8a9e\u53e5 S \u7684\u975e\u76f8\u95dc\u6a21\u578b P(w|NR S ) \u4fbf\u53ef\u7531\u6700\u4f4e\u6392\u5e8f\u6587\u4ef6\u4f86\u4f30\u6e2c\uff0c\u800c\u975e\u76f8\u95dc\u6027\u8cc7\u8a0a\u91cf\u503c\u53ef\u7531\u4e0b\u9762\u5f0f\u5b50\u8868\u793a\uff1a \uf028 \uf029 \uf028 \uf029, , m S m NR D NR KL D S M \uf03d",
                        "eq_num": "(7"
                    }
                ],
                "section": "",
                "sec_num": null
            },
            {
                "text": "\uf028 \uf029 \uf028 \uf029 \uf028 \uf029 \uf05b \uf05d, || || 2 1 min P j m m j D m Diversity D D KL D D KL D M j \uf02b \uf0d7 \uf03d \uf0ceD (8) 3.1.3\u3001\u5bc6\u5ea6\u6027\u91cf\u503c \u53e6\u4e00\u65b9\u9762\uff0c\u6700\u9ad8\u6392\u5e8f\u6587\u4ef6\u4e2d\u7684\u7d50\u69cb(Structural)\u8cc7\u8a0a\u53ef\u4ee5\u88ab\u7576\u6210\u662f\u4e00\u500b\u7dda\u7d22\u4f86\u5e6b\u52a9\u9078\u53d6\u4ee3 \u8868\u6027\u6587\u4ef6\uff0c\u5176\u4e3b\u8981\u76ee\u7684\u662f\u5e0c\u671b\u5728\u8003\u91cf\u591a\u5143\u6027\u8cc7\u8a0a\u7684\u540c\u6642\uff0c\u4e5f\u61c9\u907f\u514d\u9078\u53d6\u5230\u904e\u5ea6\u6975\u7aef\u7684\u6587 \u4ef6\uff0c\u56e0\u70ba\u904e\u5ea6\u6975\u7aef\u6587\u4ef6\u5f88\u53ef\u80fd\u662f\u932f\u8aa4\u7684\u8cc7\u8a0a\u3002\u70ba\u4e86\u5be6\u73fe\u9019\u500b\u60f3\u6cd5\uff0c\u6211\u5011\u53ef\u4ee5\u5229\u7528\u8a08\u7b97\u6700 \u9ad8\u6392\u5e8f\u6587\u4ef6\u4e2d\u7684\u5019\u9078\u6587\u4ef6 D m \u8207\u5176\u4ed6\u5019\u9078\u6587\u4ef6 D h \u7684\u8ca0\u5e73\u5747\u5c0d\u7a31(Negative Average Symmetric)\u96e2\u6563\u5ea6\u91cf\u503c\u4f86\u9054\u6210\uff0c\u5176\u516c\u5f0f\u5982\u4e0b\u6240\u793a\uff1a \uf028 \uf029 \uf028 \uf029 \uf028 \uf029 \uf05b \uf05d, || || 1 1 Top Top \uf0e5 \uf02b \uf0d7 \uf02d \uf02d \uf03d \uf0b9 \uf0ce m h h D D D h m m h m Density D D KL D D KL D M D D (9) \u5176\u4e2d|D Top |\u70ba\u6700\u9ad8\u6392\u5e8f\u6587\u4ef6\u4e4b\u500b\u6578\u3002\u82e5\u8ca0\u5e73\u5747\u5c0d\u7a31\u96e2\u6563\u5ea6\u91cf\u503c\u8d8a\u5927\uff0c\u8868\u793a\u6b64\u5019\u9078\u6587\u4ef6 D m \u8207\u6700\u9ad8\u6392\u5e8f\u6587\u4ef6\u4e2d\u7684\u5176\u4ed6\u5019\u9078\u6587\u4ef6 D h \u5f88\u63a5\u8fd1\uff0c\u4ea6\u5373\u53ef\u80fd\u662f\u6700\u9ad8\u6392\u5e8f\u6587\u4ef6\u4e2d\u5fc3\u9ede(\u5bc6\u5ea6\u5927 \u7684\u6587\u4ef6)\uff0c\u6240\u4ee5\u6b64\u5019\u9078\u6587\u4ef6 D m \u53ef\u80fd\u662f\u500b\u91cd\u8981\u7684\u6587\u4ef6\uff0c\u61c9\u8a72\u8981\u88ab\u9078\u5165\uff1b\u53cd\u4e4b\uff0c\u5c31\u6703\u96e2\u4e2d\u5fc3 \u9ede\u8f03\u9060(\u5bc6\u5ea6\u8f03\u4f4e)\uff0c\u6709\u53ef\u80fd\u6703\u662f\u4e0d\u91cd\u8981\u7684\u6587\u4ef6\uff0c\u5c31\u4e0d\u61c9\u8a72\u88ab\u9078\u5165\u3002 3.2\u3001\u91cd\u758a\u5206\u7fa4 \u672c\u8ad6\u6587\u63d0\u51fa\u4f7f\u7528\u91cd\u758a\u5206\u7fa4(Overlapped Cluster)\u7684\u6982\u5ff5\u4f86\u91cd\u65b0\u9078\u53d6\u66f4\u597d\u7684\u865b\u64ec\u76f8\u95dc\u6587\u4ef6 (\u5373 D P )\u4ee5\u5229\u63a5\u4e0b\u4f86\u5404\u7a2e\u4e0d\u540c\u7684\u95dc\u806f\u6a21\u578b\u4f30\u6e2c\u3002\u5176\u91cd\u758a\u5206\u7fa4\u9078\u53d6\u65b9\u6cd5\u70ba\u4e00\u500b\u4e09\u6b65\u9a5f\u7684\u6f14\u7b97 \u6cd5\uff0c\u793a\u610f\u5716\u5982\u5716\u4e00\u6240\u793a\uff0c\u5176\u6f14\u7b97\u6cd5\u63cf\u8ff0\u5982\u4e0b\uff1a 1. \u7b2c\u4e00\u6b65\u9a5f\uff1a\u8a08\u7b97\u6700\u9ad8\u6392\u5e8f\u6587\u4ef6\u4e2d\u5169\u5169\u5019\u9078\u6587\u4ef6\u7684\u76f8\u4f3c\u5ea6\uff0c\u5728\u6b64\u662f\u5c07\u5019\u9078\u6587\u4ef6\u8868\u9054\u6210 \u5411\u91cf\u7a7a\u9593\u6a21\u578b(Vector Space Model)\u4e26\u4f7f\u7528\u9918\u5f26\u76f8\u4f3c\u5ea6(Cosine Similarity)\u91cf\u503c\u4f86\u505a\u8a08\u7b97\u3002 2. \u7b2c\u4e8c\u6b65\u9a5f\uff1a\u5229\u7528 k-\u6700\u8fd1\u9130\u5c45(k-NN)\u4f86\u70ba\u6bcf\u500b\u5019\u9078\u6587\u4ef6 D m \u627e\u51fa k \u500b\u6700\u63a5\u8fd1\u7684\u76f8\u95dc\u6587 \u4ef6\uff0c\u4e26\u5f62\u6210\u4e00\u500b\u7fa4(\u6bcf\u500b\u5019\u9078\u6587\u4ef6\u90fd\u6703\u5f62\u6210\u4e00\u500b\u5206\u7fa4\uff0c\u800c\u6b64\u5206\u7fa4\u4e2d\u6703\u6709 k+1 \u500b\u6587\u4ef6)\u3002 3. \u7b2c\u4e09\u6b65\u9a5f\uff1a\u5c0d\u65bc\u6bcf\u500b\u5019\u9078\u6587\u4ef6 D m \u90fd\u53bb\u8a08\u7b97\u91cd\u758a\u5206\u7fa4\u7684\u500b\u6578(\u4ee5\u5716\u4e00\u4f8b\u5b50\u4f86\u8aaa\u660e\uff0c\u5019 \u9078\u6587\u4ef6\u7de8\u865f 5 \u88ab\u4e09\u500b\u5206\u7fa4\u6240\u5305\u570d\uff0c\u6240\u4ee5\u5176\u91cd\u758a\u5206\u7fa4\u500b\u6578\u70ba 3)\uff0c\u4e26\u4e14\u6309\u7167\u91cd\u758a\u5206\u7fa4\u500b\u6578\u4f86 \u5c0d\u6bcf\u500b\u5019\u9078\u6587\u4ef6\u4f5c\u6392\u5e8f\uff0c\u6392\u5e8f\u5f8c\u5373\u53ef\u5f97\u5230\u65b0\u7684\u6700\u9ad8\u6392\u5e8f\u6587\u4ef6\u3002 \u91cd\u758a\u5206\u7fa4\u7684\u6982\u5ff5\u5728\u8cc7\u8a0a\u6aa2\u7d22\u9818\u57df\u4e2d\u5df2\u6709\u4e9b\u8a31\u7814\u7a76[18]\uff0c\u5b83\u662f\u5229\u7528\u6700\u9ad8\u6392\u5e8f\u6587\u4ef6\u7684\u7d50 \u69cb\u8cc7\u8a0a\u4f86\u5e6b\u52a9\u8a13\u7df4\u8a9e\u8a00\u6a21\u578b\uff0c\u800c\u672c\u8ad6\u6587\u7684\u601d\u60f3\u662f\u8981\u5229\u7528\u91cd\u758a\u5206\u7fa4\u7684\u6982\u5ff5\u4f86\u627e\u51fa\u652f\u914d (Dominate)\u6587\u4ef6\uff0c\u82e5\u4e00\u500b\u5019\u9078\u6587\u4ef6\u7684\u91cd\u758a\u5206\u7fa4\u500b\u6578\u5f88\u591a\u7684\u8a71\uff0c\u8868\u793a\u5b83\u662f\u5f88\u91cd\u8981\u7684\u4e14\u80fd\u5920 \u652f\u914d\u5176\u4ed6\u5019\u9078\u6587\u4ef6\uff0c\u5247\u61c9\u8a72\u8981\u88ab\u9078\u70ba\u4ee3\u8868\u6027\u6587\u4ef6\u3002\u672c\u8ad6\u6587\u662f\u9996\u6b21\u5c07\u91cd\u758a\u5206\u7fa4\u7684\u6982\u5ff5\u7528\u5728 (\u8a9e\u97f3)\u6587\u4ef6\u6458\u8981\u4efb\u52d9\u4e0a\u3002 \u5716\u4e00\u3001\u4f7f\u7528\u91cd\u758a\u5206\u7fa4\u6982\u5ff5\u7684\u5019\u9078\u6587\u4ef6\u9078\u53d6\u793a\u610f\u5716\uff0c (a)\u5229\u7528 k-NN \u70ba\u6bcf\u500b\u5019\u9078\u6587\u4ef6 D m \u627e\u51fa\u91cd\u758a\u5206\u7fa4\uff0c\u4e26\u8a08\u7b97\u5176\u91cd\u758a\u5206\u7fa4\u500b\u6578\u3002 (b)\u4f9d\u64da\u6bcf\u4e00\u5019\u9078\u6587\u4ef6 D m \u7684\u91cd\u758a\u5206\u7fa4\u500b\u6578\u505a\u6392\u5e8f\u3002",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            },
            {
                "text": "\u95dc\u806f\u6a21\u578b\u7684\u57fa\u672c\u5047\u8a2d\u662f\u8a8d\u70ba\u6bcf\u4e00\u8a9e\u53e5 S \u7686\u662f\u88ab\u7528\u4f86\u63cf\u8ff0\u4e00\u500b\u6982\u5ff5\u3001\u60f3\u6cd5\u6216\u4e3b\u984c\uff0c\u6211\u5011\u7a31 \u4e4b\u70ba\u8a9e\u53e5\u7684\u95dc\u806f\u985e\u5225(Relevance Class) ",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "4.1\u3001\u95dc\u806f\u6a21\u578b",
                "sec_num": null
            },
            {
                "text": "EQUATION",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [
                    {
                        "start": 0,
                        "end": 8,
                        "text": "EQUATION",
                        "ref_id": "EQREF",
                        "raw_str": "R S \u3002\u5728\u672c\u8ad6\u6587\u4e2d\uff0c\u6211\u5011\u7684\u76ee\u6a19\u662f\u60f3\u9032\u4e00\u6b65\u5730\u6a21\u578b \u5316\u95dc\u806f\u985e\u5225\u6240\u4ee3\u8868\u7684\u8cc7\u8a0a\uff0c\u85c9\u6b64\u4f86\u8c50\u5bcc\u8a9e\u53e5\u6a21\u578b\u6240\u80fd\u50b3\u9054\u7684\u8a9e\u610f\u5167\u5bb9\u6216\u4e3b\u984c\u7279\u6027\u3002\u7136 \u800c\uff0c\u5be6\u969b\u4e0a\u6bcf\u4e00\u8a9e\u53e5 S \u7684\u95dc\u806f\u985e\u5225 S R \u662f\u975e\u5e38\u96e3\u4ee5\u6c42\u5f97\u7684\uff1b\u70ba\u6b64\uff0c\u6211\u5011\u900f\u904e\u865b\u64ec\u76f8\u95dc\u56de \u994b(Pseudo Relevant Feedback)\u4f86\u5c0b\u627e\u8207\u95dc\u806f\u985e\u5225\u53ef\u80fd\u76f8\u95dc\u7684\u4e00\u4e9b\u6587\u4ef6\uff0c\u4e26\u85c9\u7531\u9019\u4e9b\u6587\u4ef6 \u4f86 \u8fd1 \u4f3c \u95dc \u806f \u985e \u5225 R S \u3002 \u66f4 \u660e \u78ba \u5730 \uff0c \u5728 \u5be6 \u4f5c \u4e0a \u6211 \u5011 \u5c07 \u865b \u64ec \u76f8 \u95dc \u6587 \u4ef6 ( \u6700 \u9ad8 \u6392 \u5e8f \u6587 \u4ef6)D Top ={D 1 ,D 2 ,\u2026,D M }\u6216\u900f\u904e\u4e0a\u4e00\u7bc0\u6240\u4ecb\u7d39\u7684\u9078\u53d6\u65b9\u6cd5\u4f86\u7522\u751f\u8f03\u4f73\u7684\u865b\u64ec\u76f8\u95dc\u6587\u4ef6 D P \u7528\u4ee5\u4ee3\u8868\u95dc\u806f\u985e\u5225 S R \u3002\u63a5\u8457\uff0c\u900f\u904e\u6aa2\u8996\u8a5e\u5f59 w \u8207\u8a9e\u53e5 S \u5728\u9019\u4e9b\u865b\u64ec\u76f8\u95dc\u6587\u4ef6\u4e2d\u540c\u6642\u51fa \u73fe\u4e4b\u95dc\u4fc2\uff0c\u53ef\u8a08\u7b97\u51fa\u8a5e\u5f59\u8207\u8a9e\u53e5\u7684\u806f\u5408\u6a5f\u7387[12]\uff1a , ) ( ) | , ( ) , ( P RM \uf0e5 \uf03d \uf0ceD m D m m D P D S w P S w P (10) \u7576\u6211\u5011\u9032\u4e00\u6b65\u5730\u5047\u8a2d\u5728\u7d66\u5b9a\u67d0\u4e00\u7bc7\u865b\u64ec\u76f8\u95dc\u6587\u4ef6\u6642\uff0c\u8a5e\u5f59\u8207\u8a9e\u53e5\u662f\u7368\u7acb\u7684\uff0c\u4e26\u4e14\u8a9e\u53e5\u5167 \u7684\u8a5e\u5f59\u4e5f\u662f\u7368\u7acb\u4e14\u4e0d\u8003\u616e\u5176\u5148\u5f8c\u6b21\u5e8f(\u5373\u6240\u8b02\u7684\u8a5e\u888b\u5047\u8a2d)\uff0c\u5247\u900f\u904e\u865b\u64ec\u76f8\u95dc\u56de\u994b\u6240\u4f30\u6e2c \u7684\u8a9e\u53e5\u6a21\u578b\u70ba\uff1a , ) ( ) | ' ' ( ) ( ) | ( ) | ' ( ) | ( ' ' ' ' ' RM P ' P m D S w m m D m S w m D P D w P D P D w P D w P S w P m m \uf0e5 \uf0d5 \uf0e5 \uf0d5 \uf03d \uf0ce \uf0ce \uf0ce \uf0ce D D",
                        "eq_num": "(11"
                    }
                ],
                "section": "4.1\u3001\u95dc\u806f\u6a21\u578b",
                "sec_num": null
            },
            {
                "text": "D V w m \uf0d7 \uf02b \uf0d7 \uf02d \uf0d7 \uf0e5 \uf0e5 \uf03d \uf0ce \uf0ce \uf061 \uf061 D D (12) \u5176\u4e2d \u03b1 \u70ba\u5e73\u8861\u53c3\u6578\uff0c\u7528\u4f86\u63a7\u5236\u6a21\u578b\u4f30\u6e2c\u6642\u662f\u8981\u6bd4\u8f03\u504f\u597d\u7c21\u55ae\u6df7\u5408\u6a21\u578b\u6216\u662f\u80cc\u666f\u8a9e\u8a00\u6a21 \u578b\uff0cc(w,D m )\u70ba\u8a5e\u5f59 w \u5728\u865b\u64ec\u76f8\u95dc\u6587\u4ef6 D m \u7684\u6b21\u6578\uff0c\u5f0f(12)\u7684\u6700\u5927\u5316\u53ef\u900f\u904e\u671f\u671b\u503c\u6700\u5927\u5316 \u8fed\u4ee3\u66f4\u65b0\u5f0f\u4f86\u9054\u6210\uff1a \u671f\u671b\u503c\u6b65\u9a5f\uff1a , ) | ( ) 1 ( ) | ( ) | ( ) ( SMM ) ( SMM ) ( BG w P S w P S w P l l l w \uf0d7 \uf02d \uf02b \uf0d7 \uf0d7 \uf03d \uf061 \uf061 \uf061 \uf074 (13) \u6700\u5927\u5316\u6b65\u9a5f\uff1a , ) , ( ) , ( ) | ( ) ( ) ( ) 1 ( \uf0e5 \uf0e5 \uf0e5 \uf0ce \uf0a2 \uf0ce \uf0a2 \uf0ce \uf02b \uf0d7 \uf0a2 \uf0a2 \uf0d7 \uf03d V",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "4.1\u3001\u95dc\u806f\u6a21\u578b",
                "sec_num": null
            },
            {
                "text": "V w m m \uf0d7 \uf02b \uf0d7 \uf02b \uf0d7 \uf02d \uf02d \uf0d7 \uf0e5 \uf0e5 \uf03d \uf0ce \uf0ce \uf06d \uf06c \uf06d \uf06c D D (15) \u5176\u4e2d \u03bb \u548c \u03bc \u70ba\u5e73\u8861\u53c3\u6578\uff0c\u7528\u4f86\u63a7\u5236\u6a21\u578b\u4f30\u6e2c\u6642\u662f\u8981\u6bd4\u8f03\u504f\u597d\u4e09\u6df7\u5408\u6a21\u578b\u6216\u6587\u4ef6\u6a21\u578b\u4ea6\u6216 \u662f\u80cc\u666f\u8a9e\u8a00\u6a21\u578b\uff0cc(w,D m )\u70ba\u8a5e\u5f59 w \u5728\u865b\u64ec\u76f8\u95dc\u6587\u4ef6 D m \u7684\u6b21\u6578\uff0c\u5f0f(15)\u7684\u6700\u5927\u5316\u53ef\u900f\u904e \u671f\u671b\u503c\u6700\u5927\u5316\u8fed\u4ee3\u66f4\u65b0\u5f0f\u4f86\u9054\u6210\uff1a \u671f\u671b\u503c\u6b65\u9a5f\uff1a , ) | ( ) | ( ) | ( ) 1 ( ) | ( ) , ( ) | ( ) | ( ) | ( ) 1 ( ) | ( ) 1 ( ) , ( TriMM , TriMM TriMM , \uf0ef \uf0ef \uf0ee \uf0ef \uf0ef \uf0ed \uf0ec \uf0d7 \uf02b \uf0d7 \uf02b \uf0d7 \uf02d \uf02d \uf0d7 \uf0d7 \uf03d \uf0d7 \uf02b \uf0d7 \uf02b \uf0d7 \uf02d \uf02d \uf0d7 \uf02d \uf02d \uf0d7 \uf03d",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "4.1\u3001\u95dc\u806f\u6a21\u578b",
                "sec_num": null
            },
            {
                "text": "\u6458 \u8981 \u7684 \u8a55 \u4f30 \u3002 \u5176 \u8a55 \u4f30 \u7684 \u5206 \u6578 \u6709 \u4e09 \u7a2e \uff0c ROUGE-1(Unigram \uff0c \u7c21 \u5beb \u70ba R-1) \u3001 ROUGE-2(Bigram\uff0c\u7c21\u5beb\u70ba R-2)\u548c ROUGE-L(Longest Common Subsequence\uff0c\u7c21\u5beb\u70ba R-L) \u5206\u6578\uff0cROUGE-1 \u662f\u8a55\u4f30\u81ea\u52d5\u6458\u8981\u7684\u8a0a\u606f\u91cf\uff0cROUGE-2 \u662f\u8a55\u4f30\u81ea\u52d5\u6458\u8981\u7684\u6d41\u66a2\u6027\uff0c ROUGE-L \u662f\u6700\u9577\u5171\u540c\u5b57\uf905\uff0c\u672c\uf941\u6587\u5e0c\u671b\u89c0\u5bdf\u6458\u8981\u7684\u6d41\u66a2\u6027\uff0c\u56e0\u6b64\uff0c\u5be6\u9a57\u6578\u64da\u4e3b\u8981\u662f\u4ee5 ROUGE-2 \u5206\u6578\u70ba\u4e3b\u3002\u672c\u8ad6\u6587\u6240\u8a2d\u5b9a\u7684\u6458\u8981\u6bd4\u4f8b\u70ba",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "4.1\u3001\u95dc\u806f\u6a21\u578b",
                "sec_num": null
            },
            {
                "text": "RM SMM TriMM R-1 R-2 R-L R-1 R-2 R-L R-1 R-2 R-L TD Top3",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "4.1\u3001\u95dc\u806f\u6a21\u578b",
                "sec_num": null
            }
        ],
        "back_matter": [],
        "bib_entries": {
            "BIBREF0": {
                "ref_id": "b0",
                "title": "Machine-made Index for Technical Literature -an Experiment",
                "authors": [
                    {
                        "first": "P",
                        "middle": [],
                        "last": "Baxendale",
                        "suffix": ""
                    }
                ],
                "year": 1958,
                "venue": "IBM Journal of Research and Development",
                "volume": "2",
                "issue": "4",
                "pages": "354--361",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "P. Baxendale , Machine-made Index for Technical Literature -an Experiment, IBM Journal of Research and Development, Vol. 2, No. 4, pp. 354-361, 1958",
                "links": null
            },
            "BIBREF1": {
                "ref_id": "b1",
                "title": "The Use of MMR Diversity-based Reranking for Reordering Documents and Producing Summaries",
                "authors": [
                    {
                        "first": "J",
                        "middle": [],
                        "last": "Carbonell",
                        "suffix": ""
                    },
                    {
                        "first": "J",
                        "middle": [],
                        "last": "Goldstein",
                        "suffix": ""
                    }
                ],
                "year": 1998,
                "venue": "Proceedings of the 21 th Annual International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR)",
                "volume": "",
                "issue": "",
                "pages": "335--336",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "J. Carbonell and J. Goldstein, The Use of MMR Diversity-based Reranking for Reordering Documents and Producing Summaries, Proceedings of the 21 th Annual International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR), pp. 335-336, 1998",
                "links": null
            },
            "BIBREF2": {
                "ref_id": "b2",
                "title": "A Probabilistic Generative Framework for Extractive Broadcast News Speech Summarization",
                "authors": [
                    {
                        "first": "Y.-T",
                        "middle": [],
                        "last": "Chen",
                        "suffix": ""
                    },
                    {
                        "first": "B",
                        "middle": [],
                        "last": "Chen",
                        "suffix": ""
                    },
                    {
                        "first": "H.-M",
                        "middle": [],
                        "last": "Wang",
                        "suffix": ""
                    }
                ],
                "year": 2009,
                "venue": "IEEE Transactions on Audio, Speech and Language Processing",
                "volume": "17",
                "issue": "1",
                "pages": "95--106",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Y.-T. Chen, B. Chen and H.-M. Wang, A Probabilistic Generative Framework for Extractive Broadcast News Speech Summarization, IEEE Transactions on Audio, Speech and Language Processing, Vol. 17, No. 1, pp. 95-106, 2009",
                "links": null
            },
            "BIBREF3": {
                "ref_id": "b3",
                "title": "Sentence Modeling for Extractive Speech Summarization",
                "authors": [
                    {
                        "first": "B",
                        "middle": [],
                        "last": "Chen",
                        "suffix": ""
                    },
                    {
                        "first": "H.-C",
                        "middle": [],
                        "last": "Chang",
                        "suffix": ""
                    },
                    {
                        "first": "K.-Y.",
                        "middle": [],
                        "last": "Chen",
                        "suffix": ""
                    }
                ],
                "year": 2013,
                "venue": "Proceedings of the International Conference on Multimedia & Expo (ICME)",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "B. Chen, H.-C. Chang, K.-Y. Chen, Sentence Modeling for Extractive Speech Summarization, Proceedings of the International Conference on Multimedia & Expo (ICME), 2013",
                "links": null
            },
            "BIBREF4": {
                "ref_id": "b4",
                "title": "Enhancing Query Formulation for Spoken Document Retrieval",
                "authors": [
                    {
                        "first": "B",
                        "middle": [],
                        "last": "Chen",
                        "suffix": ""
                    },
                    {
                        "first": "Y.-W",
                        "middle": [],
                        "last": "Chen",
                        "suffix": ""
                    },
                    {
                        "first": "K.-Y",
                        "middle": [],
                        "last": "Chen",
                        "suffix": ""
                    }
                ],
                "year": 2014,
                "venue": "Journal of Information Science and Engineering",
                "volume": "30",
                "issue": "3",
                "pages": "553--569",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "B. Chen, Y.-W. Chen and K.-Y Chen, Enhancing Query Formulation for Spoken Document Retrieval, Journal of Information Science and Engineering, Vol. 30, No. 3, pp. 553-569, 2014",
                "links": null
            },
            "BIBREF5": {
                "ref_id": "b5",
                "title": "Text Summarization via Hidden Markov Models",
                "authors": [
                    {
                        "first": "J.-M",
                        "middle": [],
                        "last": "Conroy",
                        "suffix": ""
                    },
                    {
                        "first": "D.-P. O'",
                        "middle": [],
                        "last": "Leary",
                        "suffix": ""
                    }
                ],
                "year": 2001,
                "venue": "Proceedings of the 24 th Annual International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR)",
                "volume": "",
                "issue": "",
                "pages": "406--407",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "J.-M. Conroy and D.-P. O'Leary, Text Summarization via Hidden Markov Models, Proceedings of the 24 th Annual International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR), pp. 406-407, 2001",
                "links": null
            },
            "BIBREF6": {
                "ref_id": "b6",
                "title": "LexRank: Graph-based Lexical Centrality as Salience in Text Summarization",
                "authors": [
                    {
                        "first": "G",
                        "middle": [],
                        "last": "Erkan",
                        "suffix": ""
                    },
                    {
                        "first": "D",
                        "middle": [
                            "R"
                        ],
                        "last": "Radev",
                        "suffix": ""
                    }
                ],
                "year": 2004,
                "venue": "Journal of Artificial Intelligent Research",
                "volume": "22",
                "issue": "1",
                "pages": "457--479",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "G. Erkan and D. R. Radev, LexRank: Graph-based Lexical Centrality as Salience in Text Summarization, Journal of Artificial Intelligent Research, Vol. 22, No. 1, pp.457-479, 2004",
                "links": null
            },
            "BIBREF7": {
                "ref_id": "b7",
                "title": "Generic Text Summarization using Relevance Measure and Latent Semantic Analysis",
                "authors": [
                    {
                        "first": "Y",
                        "middle": [],
                        "last": "Gong",
                        "suffix": ""
                    },
                    {
                        "first": "X",
                        "middle": [],
                        "last": "Liu",
                        "suffix": ""
                    }
                ],
                "year": 2001,
                "venue": "Proceedings of the International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR)",
                "volume": "",
                "issue": "",
                "pages": "19--25",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Y. Gong and X. Liu, Generic Text Summarization using Relevance Measure and Latent Semantic Analysis, Proceedings of the International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR), pp. 19-25, 2001",
                "links": null
            },
            "BIBREF8": {
                "ref_id": "b8",
                "title": "Parsimonious Language Models for Information Retrieval",
                "authors": [
                    {
                        "first": "D",
                        "middle": [],
                        "last": "Hiemstra",
                        "suffix": ""
                    },
                    {
                        "first": "S",
                        "middle": [],
                        "last": "Robertson",
                        "suffix": ""
                    },
                    {
                        "first": "H",
                        "middle": [],
                        "last": "Zaragoza",
                        "suffix": ""
                    }
                ],
                "year": 2004,
                "venue": "Proceedings of the international ACM SIGIR conference on Research and development in information retrieval (SIGIR)",
                "volume": "",
                "issue": "",
                "pages": "178--185",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "D. Hiemstra, S. Robertson, and H. Zaragoza, Parsimonious Language Models for Information Retrieval, Proceedings of the international ACM SIGIR conference on Research and development in information retrieval (SIGIR), pp. 178-185, 2004",
                "links": null
            },
            "BIBREF9": {
                "ref_id": "b9",
                "title": "Summarization as Feature Selection for Text Categorization",
                "authors": [
                    {
                        "first": "A",
                        "middle": [],
                        "last": "Kolcz",
                        "suffix": ""
                    },
                    {
                        "first": "V",
                        "middle": [],
                        "last": "Prabakarmurthi",
                        "suffix": ""
                    },
                    {
                        "first": "J",
                        "middle": [],
                        "last": "Kalita",
                        "suffix": ""
                    }
                ],
                "year": 2001,
                "venue": "Proceedings of the International Conference on Information and Knowledge Management (CIKM)",
                "volume": "",
                "issue": "",
                "pages": "365--370",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "A. Kolcz, V. Prabakarmurthi and J. Kalita, Summarization as Feature Selection for Text Categorization, Proceedings of the International Conference on Information and Knowledge Management (CIKM), pp. 365-370, 2001",
                "links": null
            },
            "BIBREF10": {
                "ref_id": "b10",
                "title": "Proceedings of the Annual International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR)",
                "authors": [
                    {
                        "first": "J",
                        "middle": [],
                        "last": "Kupiec",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Trainable Document",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Summarizer",
                        "suffix": ""
                    }
                ],
                "year": 1995,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "68--73",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "J. Kupiec , A Trainable Document Summarizer, Proceedings of the Annual International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR), pp. 68-73, 1995",
                "links": null
            },
            "BIBREF11": {
                "ref_id": "b11",
                "title": "Relevance -based Language Models",
                "authors": [
                    {
                        "first": "V",
                        "middle": [],
                        "last": "Lavrenko",
                        "suffix": ""
                    },
                    {
                        "first": "W.-B",
                        "middle": [],
                        "last": "Croft",
                        "suffix": ""
                    }
                ],
                "year": 2001,
                "venue": "Proceedings of the 24 th Annual International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR)",
                "volume": "",
                "issue": "",
                "pages": "120--127",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "V. Lavrenko and W.-B. Croft, Relevance -based Language Models, Proceedings of the 24 th Annual International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR), pp. 120-127, 2001",
                "links": null
            },
            "BIBREF12": {
                "ref_id": "b12",
                "title": "Improved Speech Summarization with Multiple-hypothesis Representations and Kullback-Leibler Divergence Measures, Proceeding of the 10 th Annual Conference of the International Speech Communication Association (Interspeech)",
                "authors": [
                    {
                        "first": "S.-H",
                        "middle": [],
                        "last": "Lin",
                        "suffix": ""
                    },
                    {
                        "first": "B",
                        "middle": [],
                        "last": "Chen",
                        "suffix": ""
                    }
                ],
                "year": 2009,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "1847--1850",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "S.-H. Lin and B. Chen, Improved Speech Summarization with Multiple-hypothesis Representations and Kullback-Leibler Divergence Measures, Proceeding of the 10 th Annual Conference of the International Speech Communication Association (Interspeech), pp. 1847-1850, 2009",
                "links": null
            },
            "BIBREF13": {
                "ref_id": "b13",
                "title": "Multi-document Summarization via Budgeted Maximization of Submodular Functions",
                "authors": [
                    {
                        "first": "H",
                        "middle": [],
                        "last": "Lin",
                        "suffix": ""
                    },
                    {
                        "first": "J",
                        "middle": [],
                        "last": "Bilmes",
                        "suffix": ""
                    }
                ],
                "year": 2010,
                "venue": "Proceeding of NAACL HLT",
                "volume": "",
                "issue": "",
                "pages": "912--920",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "H. Lin and J. Bilmes, Multi-document Summarization via Budgeted Maximization of Submodular Functions, Proceeding of NAACL HLT, pp. 912-920, 2010",
                "links": null
            },
            "BIBREF14": {
                "ref_id": "b14",
                "title": "Leveraging Kullback-Leibler Divergence Measures and Information-rich Cues for Speech Summarization",
                "authors": [
                    {
                        "first": "S.-H",
                        "middle": [],
                        "last": "Lin",
                        "suffix": ""
                    },
                    {
                        "first": "Y.-M",
                        "middle": [],
                        "last": "Yeh",
                        "suffix": ""
                    },
                    {
                        "first": "B",
                        "middle": [],
                        "last": "Chen",
                        "suffix": ""
                    }
                ],
                "year": 2011,
                "venue": "IEEE Transactions on Audio, Speech and Language Processing",
                "volume": "19",
                "issue": "4",
                "pages": "871--882",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "S.-H. Lin, Y.-M. Yeh and B. Chen, Leveraging Kullback-Leibler Divergence Measures and Information-rich Cues for Speech Summarization, IEEE Transactions on Audio, Speech and Language Processing. Vol. 19, No. 4, pp. 871-882, 2011",
                "links": null
            },
            "BIBREF15": {
                "ref_id": "b15",
                "title": "ROUGE: Recall-oriented Understudy for Gisting Evaluation",
                "authors": [
                    {
                        "first": "C.-Y.",
                        "middle": [],
                        "last": "Lin",
                        "suffix": ""
                    }
                ],
                "year": 2003,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "C.-Y. Lin, ROUGE: Recall-oriented Understudy for Gisting Evaluation. 2003 [Online]. Available: http://haydn.isi.edu/ROUGE/.",
                "links": null
            },
            "BIBREF16": {
                "ref_id": "b16",
                "title": "Spoken Language Understanding: Systems for Extracting Semantic Information from Speech",
                "authors": [
                    {
                        "first": "Y",
                        "middle": [],
                        "last": "Liu",
                        "suffix": ""
                    },
                    {
                        "first": "D",
                        "middle": [],
                        "last": "Hakkani-Tur",
                        "suffix": ""
                    }
                ],
                "year": 2011,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Y. Liu and D. Hakkani-Tur, Speech Summarization, in G. Turand R. D. Mori [Ed], Spoken Language Understanding: Systems for Extracting Semantic Information from Speech, Wiley, 2011",
                "links": null
            },
            "BIBREF17": {
                "ref_id": "b17",
                "title": "Cluster-based Retrieval Using Language Models",
                "authors": [
                    {
                        "first": "X",
                        "middle": [],
                        "last": "Liu",
                        "suffix": ""
                    },
                    {
                        "first": "W",
                        "middle": [
                            "B"
                        ],
                        "last": "Croft",
                        "suffix": ""
                    }
                ],
                "year": 2004,
                "venue": "Proceedings of the International ACM SIGIR conference on Research and Development in Information Retrieval (SIGIR)",
                "volume": "",
                "issue": "",
                "pages": "186--193",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "X. Liu and W. B. Croft, Cluster-based Retrieval Using Language Models, Proceedings of the International ACM SIGIR conference on Research and Development in Information Retrieval (SIGIR), pp. 186-193, 2004",
                "links": null
            },
            "BIBREF18": {
                "ref_id": "b18",
                "title": "Effective Pseudo-relevance Feedback for Language Modeling in Extractive Speech Summarization",
                "authors": [
                    {
                        "first": "S.-H",
                        "middle": [],
                        "last": "Liu",
                        "suffix": ""
                    },
                    {
                        "first": "K.-Y",
                        "middle": [],
                        "last": "Chen",
                        "suffix": ""
                    },
                    {
                        "first": "Y.-L",
                        "middle": [],
                        "last": "Hsieh",
                        "suffix": ""
                    },
                    {
                        "first": "B",
                        "middle": [],
                        "last": "Chen",
                        "suffix": ""
                    },
                    {
                        "first": "H.-M",
                        "middle": [],
                        "last": "Wang",
                        "suffix": ""
                    },
                    {
                        "first": "H.-C",
                        "middle": [],
                        "last": "Yen",
                        "suffix": ""
                    },
                    {
                        "first": "W.-L",
                        "middle": [],
                        "last": "Hsu",
                        "suffix": ""
                    }
                ],
                "year": 2014,
                "venue": "Proceedings of the IEEE International Conference on Acoustics, Speech, and Signal Processing",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "S.-H. Liu, K.-Y. Chen, Y.-L. Hsieh, B. Chen, H.-M. Wang, H.-C. Yen, W.-L. Hsu, Effective Pseudo-relevance Feedback for Language Modeling in Extractive Speech Summarization, Proceedings of the IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP), 2014",
                "links": null
            },
            "BIBREF19": {
                "ref_id": "b19",
                "title": "Improving Sentence Modeling Techniques for Extractive Speech Summarization",
                "authors": [
                    {
                        "first": "S.-H",
                        "middle": [],
                        "last": "Liu",
                        "suffix": ""
                    },
                    {
                        "first": "K.-Y",
                        "middle": [],
                        "last": "Chen",
                        "suffix": ""
                    },
                    {
                        "first": "B",
                        "middle": [],
                        "last": "Chen",
                        "suffix": ""
                    },
                    {
                        "first": "H.-M",
                        "middle": [],
                        "last": "Wang",
                        "suffix": ""
                    },
                    {
                        "first": "W.-L",
                        "middle": [],
                        "last": "Hsu",
                        "suffix": ""
                    }
                ],
                "year": 2013,
                "venue": "ROCLING XXV: Conference on Computational Linguistics and Speech Processing",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "S.-H. Liu, K.-Y. Chen, B. Chen, H.-M. Wang, W.-L. Hsu, Improving Sentence Modeling Techniques for Extractive Speech Summarization, ROCLING XXV: Conference on Computational Linguistics and Speech Processing, 2013",
                "links": null
            },
            "BIBREF20": {
                "ref_id": "b20",
                "title": "Advances in Automatic Text Summarization",
                "authors": [
                    {
                        "first": "I",
                        "middle": [],
                        "last": "Mani",
                        "suffix": ""
                    },
                    {
                        "first": "M. -T",
                        "middle": [],
                        "last": "Maybury",
                        "suffix": ""
                    }
                ],
                "year": 1999,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "I. Mani and M. -T. Maybury, Advances in Automatic Text Summarization, Cambridge: MIT Press, 1999",
                "links": null
            },
            "BIBREF21": {
                "ref_id": "b21",
                "title": "A Study of Global Inference Algorithms in Multi-document Summarization",
                "authors": [
                    {
                        "first": "R",
                        "middle": [],
                        "last": "Mcdonald",
                        "suffix": ""
                    }
                ],
                "year": 2007,
                "venue": "Proceedings of European Conference on Information Retrieval (ECIR)",
                "volume": "",
                "issue": "",
                "pages": "557--564",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "R. McDonald, A Study of Global Inference Algorithms in Multi-document Summarization, Proceedings of European Conference on Information Retrieval (ECIR), pp. 557-564, 2007.",
                "links": null
            },
            "BIBREF22": {
                "ref_id": "b22",
                "title": "TextRank Bringing Order into Texts",
                "authors": [
                    {
                        "first": "R",
                        "middle": [],
                        "last": "Mihalcea",
                        "suffix": ""
                    },
                    {
                        "first": "P",
                        "middle": [],
                        "last": "Tarau",
                        "suffix": ""
                    }
                ],
                "year": 2004,
                "venue": "Proceedings of Empirical Method in Natural Language Processing (EMNLP)",
                "volume": "",
                "issue": "",
                "pages": "404--411",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "R. Mihalcea and P. Tarau, TextRank Bringing Order into Texts, Proceedings of Empirical Method in Natural Language Processing (EMNLP), pp. 404-411, 2004",
                "links": null
            },
            "BIBREF23": {
                "ref_id": "b23",
                "title": "Extractive Summarization of Meeting Recordings, Proceedings of the Conference of the International Speech Communication Association (Interspeech)",
                "authors": [
                    {
                        "first": "G",
                        "middle": [],
                        "last": "Murray",
                        "suffix": ""
                    },
                    {
                        "first": "S",
                        "middle": [],
                        "last": "Renals",
                        "suffix": ""
                    },
                    {
                        "first": "J",
                        "middle": [],
                        "last": "Carletta",
                        "suffix": ""
                    }
                ],
                "year": 2005,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "593--596",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "G. Murray, S. Renals, and J. Carletta, Extractive Summarization of Meeting Recordings, Proceedings of the Conference of the International Speech Communication Association (Interspeech), pp. 593-596, 2005",
                "links": null
            },
            "BIBREF24": {
                "ref_id": "b24",
                "title": "Automatic Summarization",
                "authors": [
                    {
                        "first": "A",
                        "middle": [],
                        "last": "Nenkova",
                        "suffix": ""
                    },
                    {
                        "first": "K",
                        "middle": [],
                        "last": "Mckeown",
                        "suffix": ""
                    }
                ],
                "year": 2011,
                "venue": "Foundations and Trends in Information Retrieval",
                "volume": "5",
                "issue": "2-3",
                "pages": "103--233",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "A. Nenkova and K. McKeown, Automatic Summarization, Foundations and Trends in Information Retrieval, Vol. 5, No. 2-3: 103-233, 2011",
                "links": null
            },
            "BIBREF25": {
                "ref_id": "b25",
                "title": "A Critical Reassessment of Evaluation Baselines for Speech Summarization",
                "authors": [
                    {
                        "first": "G",
                        "middle": [],
                        "last": "Penn",
                        "suffix": ""
                    },
                    {
                        "first": "X",
                        "middle": [],
                        "last": "Zhu",
                        "suffix": ""
                    }
                ],
                "year": 2008,
                "venue": "Proceedings of Annual Meeting of the Association for Computational Linguistics",
                "volume": "",
                "issue": "",
                "pages": "470--478",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "G. Penn and X. Zhu, A Critical Reassessment of Evaluation Baselines for Speech Summarization, Proceedings of Annual Meeting of the Association for Computational Linguistics, pp. 470-478, 2008",
                "links": null
            },
            "BIBREF26": {
                "ref_id": "b26",
                "title": "Active Feedback in Ad Hoc Information Retrieval",
                "authors": [
                    {
                        "first": "X",
                        "middle": [],
                        "last": "Shen",
                        "suffix": ""
                    },
                    {
                        "first": "C",
                        "middle": [],
                        "last": "Zhai",
                        "suffix": ""
                    }
                ],
                "year": 2005,
                "venue": "Proceedings of ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR)",
                "volume": "",
                "issue": "",
                "pages": "55--66",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "X. Shen and C. Zhai, Active Feedback in Ad Hoc Information Retrieval, Proceedings of ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR), pp. 55-66, 2005",
                "links": null
            },
            "BIBREF27": {
                "ref_id": "b27",
                "title": "Multi-document Summarization via the Minimum Dominating Set",
                "authors": [
                    {
                        "first": "C",
                        "middle": [],
                        "last": "Shen",
                        "suffix": ""
                    },
                    {
                        "first": "T",
                        "middle": [],
                        "last": "Li",
                        "suffix": ""
                    }
                ],
                "year": 2010,
                "venue": "Proceedings of the International Conference on Computational Linguistics (COLING)",
                "volume": "",
                "issue": "",
                "pages": "984--92",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "C. Shen and T. Li, Multi-document Summarization via the Minimum Dominating Set, Proceedings of the International Conference on Computational Linguistics (COLING), pp. 984-92, 2010",
                "links": null
            },
            "BIBREF28": {
                "ref_id": "b28",
                "title": "Document Summarization using Conditional Random Fields",
                "authors": [
                    {
                        "first": "D",
                        "middle": [],
                        "last": "Shen",
                        "suffix": ""
                    },
                    {
                        "first": "J.-T",
                        "middle": [],
                        "last": "Sun",
                        "suffix": ""
                    },
                    {
                        "first": "H",
                        "middle": [],
                        "last": "Li",
                        "suffix": ""
                    },
                    {
                        "first": "Q",
                        "middle": [],
                        "last": "Yang",
                        "suffix": ""
                    },
                    {
                        "first": "Z",
                        "middle": [],
                        "last": "Chen",
                        "suffix": ""
                    }
                ],
                "year": 2007,
                "venue": "Proceedings of International Joint Conference on Artificial Intelligence (IJCAI)",
                "volume": "",
                "issue": "",
                "pages": "2862--2867",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "D. Shen, J.-T. Sun, H. Li, Q. Yang, and Z. Chen, Document Summarization using Conditional Random Fields, Proceedings of International Joint Conference on Artificial Intelligence (IJCAI), pp. 2862-2867, 2007",
                "links": null
            },
            "BIBREF29": {
                "ref_id": "b29",
                "title": "Multi-document Summarization using Cluster-based Link Analysis",
                "authors": [
                    {
                        "first": "X",
                        "middle": [],
                        "last": "Wan",
                        "suffix": ""
                    },
                    {
                        "first": "J",
                        "middle": [],
                        "last": "Yang",
                        "suffix": ""
                    }
                ],
                "year": 2008,
                "venue": "Proceedings of the Annual International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR)",
                "volume": "",
                "issue": "",
                "pages": "299--306",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "X. Wan and J. Yang, Multi-document Summarization using Cluster-based Link Analysis, Proceedings of the Annual International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR), pp. 299-306, 2008",
                "links": null
            },
            "BIBREF30": {
                "ref_id": "b30",
                "title": "Incorporating Diversity and Density in Active Learning for Relevance Feedback",
                "authors": [
                    {
                        "first": "Z",
                        "middle": [],
                        "last": "Xu",
                        "suffix": ""
                    },
                    {
                        "first": "R",
                        "middle": [],
                        "last": "Akella",
                        "suffix": ""
                    },
                    {
                        "first": "Y",
                        "middle": [],
                        "last": "Zhang",
                        "suffix": ""
                    }
                ],
                "year": 2007,
                "venue": "Proceedings of European Conference on Information Retrieval (ECIR)",
                "volume": "",
                "issue": "",
                "pages": "245--257",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Z. Xu, R. Akella and Y. Zhang, Incorporating Diversity and Density in Active Learning for Relevance Feedback, Proceedings of European Conference on Information Retrieval (ECIR), pp. 245-257, 2007",
                "links": null
            },
            "BIBREF31": {
                "ref_id": "b31",
                "title": "Model-based feedback in the language modeling approach to information retrieval",
                "authors": [
                    {
                        "first": "C.-X",
                        "middle": [],
                        "last": "Zhai",
                        "suffix": ""
                    },
                    {
                        "first": "J",
                        "middle": [],
                        "last": "Lafferty",
                        "suffix": ""
                    }
                ],
                "year": 2001,
                "venue": "Proceeding of the International Conference on Information and Knowledge Management (CIKM)",
                "volume": "",
                "issue": "",
                "pages": "403--410",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "C.-X. Zhai and J. Lafferty, Model-based feedback in the language modeling approach to information retrieval, Proceeding of the International Conference on Information and Knowledge Management (CIKM), pp. 403-410, 2001",
                "links": null
            },
            "BIBREF32": {
                "ref_id": "b32",
                "title": "A Study of Smoothing Methods for Language Models Applied to Ad Hoc Information Retrieval",
                "authors": [
                    {
                        "first": "C.-X",
                        "middle": [],
                        "last": "Zhai",
                        "suffix": ""
                    },
                    {
                        "first": "J",
                        "middle": [],
                        "last": "Lafferty",
                        "suffix": ""
                    }
                ],
                "year": 2011,
                "venue": "Proceedings of the Annual International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR)",
                "volume": "",
                "issue": "",
                "pages": "334--342",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "C.-X. Zhai and J. Lafferty, A Study of Smoothing Methods for Language Models Applied to Ad Hoc Information Retrieval, Proceedings of the Annual International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR), pp. 334-342, 2011",
                "links": null
            },
            "BIBREF33": {
                "ref_id": "b33",
                "title": "Statistical Language Models for Information Retrieval: A Critical Review",
                "authors": [
                    {
                        "first": "C.-X",
                        "middle": [],
                        "last": "Zhai",
                        "suffix": ""
                    }
                ],
                "year": 2008,
                "venue": "Foundations and Trends in Information Retrieval",
                "volume": "2",
                "issue": "3",
                "pages": "137--213",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "C.-X. Zhai, Statistical Language Models for Information Retrieval: A Critical Review, Foundations and Trends in Information Retrieval, Vol. 2, No.3, pp.137-213, 2008",
                "links": null
            },
            "BIBREF34": {
                "ref_id": "b34",
                "title": "Speech Summarization without Lexical Features for Mandarin Broadcast News",
                "authors": [
                    {
                        "first": "J",
                        "middle": [],
                        "last": "Zhang",
                        "suffix": ""
                    },
                    {
                        "first": "P",
                        "middle": [],
                        "last": "Fung",
                        "suffix": ""
                    }
                ],
                "year": 2007,
                "venue": "Proceedings of NAACL HLT, Companion Volume",
                "volume": "",
                "issue": "",
                "pages": "213--216",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "J. Zhang and P. Fung, Speech Summarization without Lexical Features for Mandarin Broadcast News, Proceedings of NAACL HLT, Companion Volume, pp. 213-216, 2007",
                "links": null
            }
        },
        "ref_entries": {
            "FIGREF0": {
                "uris": null,
                "num": null,
                "type_str": "figure",
                "text": "D Top )\uff0c\u6216\u9032\u4e00\u6b65\u5730\u4f7f \u7528\u4e0a\u4e00\u7bc0\u63d0\u51fa\u4e4b\u65b9\u6cd5\u4f86\u6539\u5584\u865b\u64ec\u76f8\u95dc\u6587\u4ef6\u5f8c(\u5373 D P )\uff0c\u63a5\u4e0b\u4f86\u5c31\u8981\u505a\u6a21\u578b\u4f30\u6e2c\uff0c\u5e95\u4e0b\u4ecb\u7d39 \u5e38\u898b\u7684\u6a21\u578b\u5305\u542b\u6709\u95dc\u806f\u6a21\u578b(Relevance Model, RM)\u3001\u7c21\u55ae\u6df7\u5408\u6a21\u578b(Simple Mixture Model, SMM)\u4ee5\u53ca\u4e09\u6df7\u5408\u6a21\u578b(Tri-Mixture Model, TriMM)\u3002"
            },
            "FIGREF1": {
                "uris": null,
                "num": null,
                "type_str": "figure",
                "text": "\u5176\u4e2d l \u8868\u793a\u671f\u671b\u503c\u6700\u5927\u5316\u7684\u7b2c l \u6b21\u8fed\u4ee3\u3002\u9019\u500b\u7c21\u55ae\u6df7\u5408\u6a21\u578b\u7684\u4f30\u6e2c\u6703\u52a0\u5f37\u5177\u6709\u7368\u7279\u6027 (Specificity)\u7684\u8a5e\u5f59\u4e4b\u6a5f\u7387\uff0c\u4f8b\u5982\u67d0\u8a5e\u5f59\u6c92\u6709\u5728\u80cc\u666f\u8a9e\u8a00\u6a21\u578b\u4e2d\u6709\u597d\u89e3\u91cb(Well-Explained) \u5247\u6703\u88ab\u52a0\u5f37\u5176\u6a5f\u7387\uff0c\u9019\u6a23\u4f7f\u5f97\u6b64\u6a21\u578b\u70ba\u66f4\u5177\u6709\u9451\u5225(Discriminant)\u80fd\u529b\u7684\u8a9e\u53e5\u6a21\u578b\uff1b\u53cd \u4e4b\uff0c\u82e5\u662f\u6c92\u6709\u7368\u7279\u6027\u7684\u8a5e\u5f59\uff0c\u5247\u5176\u6a5f\u7387\u5c31\u6703\u88ab\u80cc\u666f\u8a9e\u8a00\u6a21\u578b\u6240\u5438\u6536\u3002"
            },
            "FIGREF2": {
                "uris": null,
                "num": null,
                "type_str": "figure",
                "text": "\u904b\u7528\u6b64\u4e09\u6df7\u5408\u6a21\u578b\u4f86\u8abf\u9069\u8a9e\u53e5\u6a21\u578b\u6642\uff0c\u5eab\u723e\u8c9d\u514b-\u840a\u4f2f\u52d2\u96e2\u6563\u5ea6\u91cf\u503c\u7684\u516c\u5f0f(\u53c3\u7167\u5f0f(5)) \u53ef\u9032\u4e00\u6b65\u5730\u8868\u793a\u6210\uff1a"
            },
            "FIGREF3": {
                "uris": null,
                "num": null,
                "type_str": "figure",
                "text": "\u81f4\u8b1d \u672c\u8ad6\u6587\u4e4b\u7814\u7a76\u627f\u8499\u6559\u80b2\u90e8-\u570b\u7acb\u81fa\u7063\u5e2b\u7bc4\u5927\u5b78\u9081\u5411\u9802\u5c16\u5927\u5b78\u8a08\u756b(102J1A0800)\u8207\u884c\u653f\u9662 \u79d1 \u6280 \u90e8 \u7814 \u7a76 \u8a08 \u756b (MOST 103-2221-E-003-016-MY2, NSC 103-2911-I-003-301, NSC 101-2221-E-003-024-MY3 \u3001 NSC 101-2511-S-003-057-MY3 \u3001 NSC 101-2511-S-003-047-MY3 \u548c NSC 102-2221-E-003-014-MY3)\u4e4b\u7d93\u8cbb\u652f\u6301\uff0c\u8b39\u6b64\u81f4\u8b1d\u3002"
            },
            "TABREF0": {
                "type_str": "table",
                "num": null,
                "text": "\u3001\u6f5b\u85cf\u8a9e\u610f\u5206\u6790(Latent Semantic Analysis, LSA)[8][13]\u53ca\u6700\u5927\u908a\u969b\u95dc\u806f(Maximal Marginal Relevance, MMR)[2]\u7b49\uff1b(II)\u4ee5\u5716(Graph)\u70ba\u57fa\u790e\u7684\u65b9\u6cd5\uff0c\u5177\u4ee3\u8868\u6027\u7684\u6709\u99ac\u53ef\u592b\u96a8 \u6a5f\u6f2b\u8d70(Markov Random Walk, MRW)[30]\u3001\u8a5e\u5f59\u6392\u5e8f(LexRank)[7][23]\u53ca\u6700\u5c0f\u652f\u914d\u96c6 (Minimal Dominating Set)[28]\u7b49\uff1b(III)\u4ee5\u7d44\u5408\u6700\u4f73(Combinatorial Optimization)\u70ba\u57fa\u790e\u7684\u65b9 \u6cd5\uff0c\u5305\u62ec\u6709\u6b21\u6a21(Submodularity)[14]\u4ee5\u53ca\u7dda\u6027\u6574\u6578\u898f\u5283(Linear Integer Programming)[22] Na\u00efve-Bayes Classifier)[11]\u3001\u9ad8\u65af\u6df7\u5408\u6a21\u578b(Gaussian Mixture Model, GMM)[24]\u3001\u96b1\u85cf \u5f0f\u99ac\u53ef\u592b\u6a21\u578b(Hidden Markov Model, HMM)[6]\u3001\u652f\u63f4\u5411\u91cf\u6a5f(Support Vector Machines, SVM)[10]\u8207\u689d\u4ef6\u96a8\u6a5f\u5834\u57df(Conditional Random Fields, CRF)[29]\u7b49\u3002\u7531\u65bc\u76e3\u7763\u5f0f\u6458\u8981\u65b9\u6cd5",
                "content": "<table><tr><td>\u6700\u5f8c\uff0c\u7b2c\u4e03\u7bc0\u70ba\u7d50\u8ad6\u8207\u672a\u4f86\u7814\u7a76\u65b9\u5411\u3002</td></tr><tr><td>\u4e8c\u3001\u4f7f\u7528\u8a9e\u8a00\u6a21\u578b\u65bc\u8a9e\u97f3\u6587\u4ef6\u6458\u8981</td></tr><tr><td>\u5728\u904e\u53bb\u8fd1\u4e8c\u5341\uf98e\u4f86\uff0c\u5404\u7a2e\u8a9e\u8a00\u6a21\u578b\u5728\u8cc7\u8a0a\u6aa2\u7d22\u4efb\u52d9\u4e2d\u5df2\u88ab\u5ee3\u6cdb\u5730\u61c9\u7528\uff0c\u4e26\u4e14\u6709\u4e0d\u932f\u7684\u5be6</td></tr><tr><td>\u52d9\u6210\u6548[34]\u3002\u8fd1\u671f\u5728\u8a9e\u97f3\u6458\u8981\u9818\u57df\uff0c\u4ea6\u958b\u59cb\u6709\u4e00\u4e9b\u57fa\u65bc\u8a9e\u8a00\u6a21\u578b\u5316\u67b6\u69cb\u7684\u975e\u76e3\u7763\u5f0f\u6458\u8981</td></tr><tr><td>\u65b9\u6cd5\u88ab\u63d0\u51fa\u3002\u672c\u7bc0\u5c07\u5148\u7c21\u4ecb\u5169\u7a2e\u5e38\u898b\u7684\u3001\u57fa\u65bc\u8a9e\u8a00\u6a21\u578b\u5316\u67b6\u69cb\u7684\u6458\u8981\u65b9\u6cd5\uff1a\u5176\u4e00\u70ba\u4f7f\u7528</td></tr><tr><td>\u8a9e\u53e5\u8a9e\u8a00\u6a21\u578b\u751f\u6210\u6587\u4ef6\u7684\u6587\u4ef6\u76f8\u4f3c\u5ea6\u91cf\u503c(Document Likelihood Measure, DLM)\u7684\u6458\u8981</td></tr><tr><td>\u65b9\u6cd5[3]\uff1b\u53e6\u5916\u70ba\u4f7f\u7528\u5eab\u723e\u8c9d\u514b-\u840a\u4f2f\u52d2\u96e2\u6563\u5ea6\u91cf\u503c(Kullback-Leibler Divergence Measure,</td></tr><tr><td>KL)[13][15]\u4f86\u8a08\u7b97\u6587\u4ef6\u6a21\u578b\u548c\u8a9e\u53e5\u6a21\u578b\u9593\u4e4b\u8ddd\u96e2\u7684\u6458\u8981\u65b9\u6cd5\u3002\u63a5\u8457\uff0c\u6211\u5011\u5c07\u95e1\u8ff0\u5982\u4f55\u5229</td></tr><tr><td>\u7528\u865b\u64ec\u76f8\u95dc\u56de\u994b(Pseudo Relevance Feedback)\u6982\u5ff5\u4f86\u7372\u5f97\u66f4\u53ef\u9760\u7684\u8a9e\u53e5\u6a21\u578b\u4f30\u6e2c\uff0c\u4e26\u4ecb\u7d39</td></tr><tr><td>\u6578\u500b\u5728\u8cc7\u8a0a\u6aa2\u7d22\u9818\u57df\u5df2\u88ab\u767c\u5c55\u51fa\u7684\u65b0\u7a4e\u865b\u64ec\u76f8\u95dc\u6587\u4ef6\u9078\u53d6\u6280\u8853\u3002</td></tr><tr><td>2.1\u3001\u6587\u4ef6\u76f8\u4f3c\u5ea6\u91cf\u503c</td></tr><tr><td>\u7b49\u3002\u6700\u5f8c\uff0c\u7b2c\u4e09\u985e\u76e3\u7763\u5f0f\u6458\u8981\u65b9\u6cd5\u901a\u5e38\u5c07\u81ea\u52d5\u6458\u8981\u4e4b\u4efb\u52d9\u8996\u70ba\u4e8c\u5143\u5206\u985e\u554f\u984c(Binary \u6211\u5011\u53ef\u4ee5\u628a\u8a9e\u97f3\u6587\u4ef6\u6458\u8981\u4efb\u52d9\u8996\u70ba\u662f\u8cc7\u8a0a\u6aa2\u7d22\u7684\u554f\u984c\u3002\u4e00\u822c\u4f86\u8aaa\uff0c\u8cc7\u8a0a\u6aa2\u7d22(Information</td></tr><tr><td>Classification)\uff0c\u4ea6\u5373\u5c07\u8a9e\u53e5\u5340\u5206\u70ba\u6458\u8981\u8a9e\u53e5\u6216\u975e\u6458\u8981\u8a9e\u53e5\u3002\u5728\u8a13\u7df4\u968e\u6bb5\uff0c\u5fc5\u9808\u4e8b\u5148\u6e96\u5099 Retrieval, IR)\u65e8\u5728\u5c0b\u627e\u76f8\u95dc\u6587\u4ef6(Relevant Document)\u4f86\u56de\u61c9\u4f7f\u7528\u8005\u6240\u9001\u51fa\u7684\u67e5\u8a62(Query)</td></tr><tr><td>\u597d\u4e00\u4e9b\u8a13\u7df4\u6587\u4ef6\u4ee5\u53ca\u5176\u5c0d\u61c9\u7684\u4eba\u5de5\u6a19\u8a3b\u904e\u6458\u8981\u8cc7\u8a0a\uff0c\u7136\u5f8c\u7d50\u5408\u5404\u7a2e\u8a5e\u5f59\u3001\u8a9e\u610f\u6216\u97f3\u97fb\u7b49 \u6216\u8cc7\u8a0a\u9700\u6c42(Information Need)\u3002\u540c\u6a23\u5730\uff0c\u5728\u5f9e\u4e8b\u8a9e\u97f3\u6587\u4ef6\u6458\u8981\u6642\uff0c\u6211\u5011\u53ef\u5c07\u6bcf\u4e00\u7bc7\u88ab\u6458</td></tr><tr><td>\u7279\u5fb5\u4f86\u8868\u793a\u8a9e\u53e5\uff0c\u4e26\u4e14\u900f\u904e\u5404\u7a2e\u5206\u985e\u5668\u7684\u5b78\u7fd2\u6a5f\u5236\u9032\u884c\u6458\u8981(\u5206\u985e)\u6a21\u578b\u7684\u8a13\u7df4\uff1b\u5728\u6e2c\u8a66 \u8981\u6587\u4ef6\u8996\u70ba\u662f\u67e5\u8a62\uff0c\u800c\u6587\u4ef6\u4e2d\u7684\u6bcf\u4e00\u53e5\u8a9e\u53e5(Sentence)\u8996\u70ba\u5019\u9078\u8cc7\u8a0a\u55ae\u5143(Candidate</td></tr><tr><td>\u968e\u6bb5\uff0c\u5c0d\u65bc\u5c07\u88ab\u6458\u8981\u4e4b\u6587\u4ef6\uff0c\u6b64\u985e\u65b9\u6cd5\u5c07\u6587\u4ef6\u88e1\u7684\u6bcf\u4e00\u53e5\u8a9e\u53e5\u9032\u884c\u4e8c\u5143\u5206\u985e\uff0c\u5373\u53ef\u4f9d\u6240 Information Unit)\uff1b\u64da\u6b64\uff0c\u6211\u5011\u53ef\u4ee5\u5047\u8a2d\u5728\u88ab\u6458\u8981\u6587\u4ef6\u4e2d\uff0c\u8207\u6587\u4ef6\u672c\u8eab\u6108\u76f8\u95dc\u7684\u8a9e\u53e5\u5c31</td></tr><tr><td>\u8a2d\u5b9a\u6458\u8981\u6bd4\u4f8b\u4f86\u53d6\u6458\u8981\u8a9e\u53e5\u4ee5\u7522\u751f\u51fa\u6458\u8981\u3002\u5728\u6b64\u985e\u65b9\u6cd5\u4e2d\uff0c\u8f03\u8457\u540d\u7684\u5305\u62ec\u7c21\u55ae\u8c9d\u6c0f\u5206\u985e \u6108\u6709\u53ef\u80fd\u662f\u53ef\u7528\u4f86\u4ee3\u8868\u6587\u4ef6\u4e3b\u65e8\u6216\u4e3b\u984c\u4e4b\u6458\u8981\u8a9e\u53e5\u3002</td></tr><tr><td>\u5668(\u6240\u4f7f\u7528\u7684\u6a21\u578b\u5728\u8a13\u7df4\u6642\u5fc5\u9808\u4f7f\u7528\u4e00\u5b9a\u6578\u91cf\u6587\u4ef6\u53ca\u5176\u5c0d\u61c9\u7d93\u4eba\u5de5\u6a19\u8a3b\u904e\u6458\u8981\u8cc7\u8a0a\uff0c\u6240\u4ee5\u7576</td></tr><tr><td>\u5b83\u5011\u88ab\u61c9\u7528\u5230\u65b0\u7684\u6458\u8981\u4efb\u52d9\u6216\u61c9\u7528\u9818\u57df\u6642\uff0c\u76f8\u8f03\u4e0a\u8ff0\u5169\u985e\u6458\u8981\u65b9\u6cd5\u800c\u8a00\uff0c\u662f\u6703\u8017\u8cbb\u8a31\u591a</td></tr><tr><td>\u4eba\u529b\u8207\u6642\u9593\u7684\u3002\u503c\u5f97\u4e00\u63d0\u7684\u662f\uff0c\u81ea\u52d5\u6458\u8981\u7814\u7a76\u4e5f\u53ef\u5f9e\u5176\u5b83\u4e0d\u540c\u9762\u76f8\u4f86\u9032\u884c\u63a2\u8a0e\uff0c\u5305\u62ec\u4e86</td></tr><tr><td>\u4f86\u6e90\u3001\u9700\u6c42\u3001\u65b9\u5f0f\u3001\u7528\u9014\u7b49\uff0c\u6709\u8208\u8da3\u7684\u8b80\u8005\u53ef\u53c3\u8003\u76f8\u95dc\u6587\u737b[17][20][21][25]\u9032\u884c\u66f4\u6df1\u5165</td></tr><tr><td>\u7684\u77ad\u89e3\u3002</td></tr><tr><td>\u6709\u5225\u65bc\u4e0a\u8ff0\u7684\u6458\u8981\u65b9\u6cd5\uff0c\u8fd1\u671f\u6709\u4e00\u4e9b\u57fa\u65bc\u8a9e\u8a00\u6a21\u578b\u5316(Language Modeling, LM)\u67b6\u69cb</td></tr><tr><td>\u4e4b\u6458\u8981\u65b9\u6cd5\u88ab\u63d0\u51fa\uff0c\u4e26\u4e14\u521d\u6b65\u5728\u7bc0\u9304\u5f0f\u6587\u5b57\u6216\u8a9e\u97f3\u6458\u8981\u4efb\u52d9\u4e0a\u5c55\u73fe\u4e0d\u932f\u7684\u6548\u80fd\u3002\u5728\u6b64\u67b6</td></tr><tr><td>\u69cb\u4e0b\uff0c\u5c0d\u65bc\u88ab\u6458\u8981\u6587\u4ef6\u6bcf\u4e00\u53e5\u5019\u9078\u8a9e\u53e5\u4e4b\u8a9e\u53e5\u6a21\u578b\u7684\u5efa\u7acb\uff0c\u53ef\u900f\u904e\u865b\u64ec\u76f8\u95dc\u56de\u994b(Pseudo</td></tr><tr><td>Relevance Feedback, PRF)\u7b56\u7565\u4f86\u7372\u5f97\u66f4\u52a0\u53ef\u9760\u7684\u53c3\u6578\u4f30\u6e2c\u3002\u4e00\u822c\u4f86\u8aaa\uff0c\u865b\u64ec\u76f8\u95dc\u56de\u994b\u5728</td></tr><tr><td>\u57f7\u884c\u4e0a\u53ef\u5206\u70ba\u5169\u500b\u968e\u6bb5\uff1a1)\u76f8\u95dc\u8cc7\u8a0a(\u6216\u8005\u660e\u78ba\u5730\u8aaa\uff0c\u865b\u64ec\u76f8\u95dc\u6587\u4ef6)\u7684\u9078\u53d6\uff1b2)\u8a9e\u53e5\u6a21</td></tr><tr><td>\u578b\u5316\u8207\u53c3\u6578\u91cd\u65b0\u4f30\u6e2c\u3002\u672c\u8ad6\u6587\u540c\u6a23\u4e5f\u57fa\u65bc\u8a9e\u8a00\u6a21\u578b\u5316\u67b6\u69cb\u4f86\u767c\u5c55\u8a9e\u97f3\u6458\u8981\u65b9\u6cd5\uff0c\u5176\u8ca2\u737b</td></tr><tr><td>\u4e3b\u8981\u6709\u4e09\u65b9\u9762\u3002\u9996\u5148\uff0c\u6709\u5225\u65bc\u73fe\u6709\u57fa\u65bc\u8a9e\u8a00\u6a21\u578b\u5316\u67b6\u69cb\u4e4b\u6458\u8981\u65b9\u6cd5\u90fd\u805a\u7126\u5728\u8a9e\u53e5\u6a21\u578b\u53c3</td></tr><tr><td>\u6578\u7684\u91cd\u65b0\u4f30\u6e2c\uff0c\u672c\u8ad6\u6587\u6df1\u5165\u63a2\u8a0e\u8207\u61c9\u7528\u5404\u7a2e\u9069\u5408\u65bc\u7bc0\u9304\u5f0f\u8a9e\u97f3\u6587\u4ef6\u6458\u8981\u7684\u865b\u64ec\u76f8\u95dc\u6587\u4ef6</td></tr><tr><td>\u9078\u53d6\u6280\u8853\uff0c\u7528\u4ee5\u5f37\u5316\u8a9e\u53e5\u6a21\u578b\u7684\u53c3\u6578\u4f30\u6e2c\u3002\u5176\u6b21\uff0c\u672c\u8ad6\u6587\u66f4\u9032\u4e00\u6b65\u5730\u8003\u91cf\u4f7f\u7528\u6bcf\u4e00\u8a9e\u53e5</td></tr><tr><td>\u7684\u975e\u76f8\u95dc\u6027(Non-relevance)\u8cc7\u8a0a\u5c0d\u65bc\u865b\u64ec\u76f8\u95dc\u6587\u4ef6\u9078\u53d6\u7684\u5f71\u97ff\u3002\u540c\u6642\uff0c\u6211\u5011\u4ea6\u984d\u5916\u5617\u8a66</td></tr><tr><td>\u57fa\u65bc\u91cd\u758a\u5206\u7fa4(Overlapped Clustering)\u6982\u5ff5\u4f86\u6709\u6548\u5730\u9078\u53d6\u91cd\u8981\u7684\u865b\u64ec\u76f8\u95dc\u6587\u4ef6\u505a\u70ba\u8a9e\u53e5</td></tr><tr><td>\u6a21\u578b\u7684\u53c3\u6578\u4f30\u6e2c\u4e4b\u4f9d\u64da\u3002\u6700\u5f8c\uff0c\u672c\u8ad6\u6587\u63a2\u7d22\u4f7f\u7528\u4e09\u6df7\u5408\u6a21\u578b(Tri-Mixture Model)\u4f86\u8868\u793a</td></tr><tr><td>\u6bcf\u4e00\u8a9e\u53e5\uff0c\u671f\u76fc\u5176\u80fd\u66f4\u7cbe\u78ba\u5730\u8868\u793a\u4e00\u53e5\u8a9e\u53e5\u4e4b\u7368\u7279\u8a5e\u5f59\u4f7f\u7528\u548c\u8a9e\u610f\u76f8\u95dc\u8cc7\u8a0a\u3002</td></tr><tr><td>\u672c\u8ad6\u6587\u5f8c\u7e8c\u5b89\u6392\u5982\u4e0b\uff1a\u7b2c\u4e8c\u7bc0\u9996\u5148\u4ecb\u7d39\u4f7f\u7528\u8a9e\u8a00\u6a21\u578b\u65bc\u7bc0\u9304\u5f0f\u8a9e\u97f3\u6458\u8981\u4efb\u52d9\u4e4b\u539f</td></tr><tr><td>\u7406\uff0c\u7136\u5f8c\u95e1\u8ff0\u865b\u64ec\u76f8\u95dc\u56de\u994b\u7684\u89c0\u5ff5\u53ca\u5176\u73fe\u6709\u865b\u64ec\u76f8\u95dc\u6587\u4ef6\u9078\u53d6\u6280\u8853\uff1b\u7b2c\u4e09\u7bc0\u5c07\u4ecb\u7d39\u672c\u8ad6</td></tr><tr><td>\u6587\u63d0\u51fa\u4e4b\u65b0\u7a4e\u5f0f\u865b\u64ec\u76f8\u95dc\u6587\u4ef6\u9078\u53d6\u6280\u8853\uff1b\u7b2c\u56db\u7bc0\u5247\u4ecb\u7d39\u73fe\u6709\u5404\u7a2e\u95dc\u806f\u6a21\u578b\uff0c\u4e26\u4e14\u8aaa\u660e\u5982</td></tr><tr><td>\u4f55\u7d50\u5408\u8a9e\u53e5\u95dc\u806f\u6027\u8cc7\u8a0a\u4f86\u6539\u9032\u8a9e\u53e5\u6a21\u578b\u4e4b\u4f30\u6e2c\uff0c\u4f7f\u5176\u5f97\u4ee5\u66f4\u7cbe\u6e96\u5730\u4ee3\u8868\u8a9e\u53e5\u7684\u8a9e\u610f\u5167</td></tr><tr><td>\u5bb9\uff1b\u7b2c\u4e94\u7bc0\u4ecb\u7d39\u5be6\u9a57\u8a9e\u6599\u8207\u8a2d\u5b9a\u4ee5\u53ca\u6458\u8981\u8a55\u4f30\u4e4b\u65b9\u6cd5\uff1b\u7b2c\u516d\u7bc0\u8aaa\u660e\u5be6\u9a57\u7d50\u679c\u53ca\u5176\u5206\u6790\uff1b</td></tr></table>",
                "html": null
            },
            "TABREF6": {
                "type_str": "table",
                "num": null,
                "text": "\u8868\u4e8c\u70ba\u672c\u8ad6\u6587\u4e4b\u57fa\u790e\u5be6\u9a57\u7d50\u679c\u3002\u9996\u5148\uff0c\u5728 TD \u7684\u5be6\u9a57\u4e2d\uff0cKL \u7684\u6458\u8981\u6548\u679c\u6bd4 LS\u3001 LEAD\u3001VSM\u3001LSA\u3001MMR \u7b49\u975e\u76e3\u7763\u5f0f\u6458\u8981\u65b9\u6cd5\u4f86\u5f97\u597d\u4e9b\uff1b\u56e0 LS \u8207 LEAD \u50c5\u9069\u7528\u65bc \u7279\u6b8a\u6587\u7ae0\u7d50\u69cb\u4e0a\uff0c\u6240\u4ee5\u82e5\u88ab\u6458\u8981\u6587\u4ef6\u4e0d\u5177\u6709\u67d0\u7a2e\u7279\u6b8a\u7684\u6587\u7ae0\u7d50\u69cb\uff0c\u5176\u6458\u8981\u6548\u80fd\u5c31\u6703\u6709 \u9650\u3002\u76f8\u8f03\u4e4b\u4e0b\uff0cKL \u662f\u8f03\u5177\u4e00\u822c\u6027\u7684\u6458\u8981\u65b9\u6cd5\uff0c\u56e0\u6b64\u6bd4\u8f03\u4e0d\u6703\u53d7\u9650\u65bc\u6587\u7ae0\u7684\u7d50\u69cb\u4e4b\u5f71\u97ff\uff0c \u6545\u6458\u8981\u6548\u80fd\u6bd4 LS \u4ee5\u53ca LEAD \u4f86\u5f97\u5f70\u986f\u3002KL \u8207 VSM \u7686\u4f7f\u7528\u6dfa\u5c64\u7684\u8a5e\u5f59(\u8a5e\u983b)\u8cc7\u8a0a\uff0c\u4f46 \u7531\u65bc KL \u662f\u8a08\u7b97\u8a9e\u53e5\u6a21\u578b\u8207\u6587\u4ef6\u6a21\u578b\u4e4b\u9593\u7684\u8ddd\u96e2\u95dc\u4fc2\uff0c\u5c0d\u65bc\u4ee3\u8868\u8a9e\u53e5\u8207\u6587\u4ef6\u7684\u8a9e\u8a00\u6a21 \u578b\uff0c\u6211\u5011\u8f03\u5bb9\u6613\u900f\u904e\u5404\u7a2e\u6280\u8853\u4f86\u9032\u884c\u6a21\u578b\u7684\u4f30\u8a08\u8207\u8abf\u9069\uff0c\u9032\u800c\u7372\u5f97\u8f03\u597d\u7684\u6458\u8981\u6210\u679c\u3002 MMR \u5728\u9078\u53d6\u6642\u591a\u8003\u616e\u4e86\u5197\u9918\u8cc7\u8a0a\uff0c\u6240\u4ee5\u6458\u8981\u6548\u679c\u4e5f\u6bd4 VSM \u4f86\u5f97\u597d\u4e9b\u3002LSA \u5728\u6f5b\u85cf\u8a9e \u610f\u7a7a\u9593\u8a08\u7b97\u6587\u4ef6\u8207\u8a9e\u53e5\u7684\u9918\u5f26\u76f8\u4f3c\u5ea6\u91cf\u503c\uff0c\u5176\u7d50\u679c\u986f\u793a\u4e5f\u6703\u8f03 VSM \u597d\u3002\u6574\u6578\u7dda\u6027\u898f\u5283 \u662f\u4e00\u500b\u5168\u57df\u9078\u64c7\u65b9\u6cd5\uff0c\u6240\u4ee5\u5728 TD \u4e0a\u53ef\u4ee5\u5f97\u5230\u6700\u597d\u7684\u6458\u8981\u6548\u80fd\u3002\u53e6\u4e00\u65b9\u9762\uff0c\u5728 SD \u7684\u5be6 \u9a57\u4e2d\uff0cKL \u540c\u6a23\u8f03\u512a\u65bc LS\u3001LEAD \u4e4b\u6458\u8981\u65b9\u6cd5\uff0c\u4f46 VSM \u7684\u7d50\u679c\u5247\u7a0d\u5fae\u8f03 KL \u597d\u4e00\u9ede\uff0c \u6211\u5011\u8a8d\u70ba\u9019\u53ef\u80fd\u662f\u56e0\u70ba VSM \u6bd4\u8f03\u4e0d\u53d7\u5230\u8a9e\u97f3\u8fa8\u8a8d\u932f\u8aa4\u7684\u5f71\u97ff\u3002\u539f\u4ee5\u70ba ILP \u4e5f\u6703\u5728 SD \u4e2d\u5f97\u5230\u6700\u597d\u7684\u6458\u8981\u6548\u80fd\uff0c\u7d50\u679c\u53cd\u800c\u662f MMR \u5f97\u5230\u6700\u597d\u7684\u6458\u8981\u6548\u80fd\uff0c\u53ef\u80fd\u7684\u539f\u56e0\u662f ILP \u53d7 \u5230\u8a9e\u97f3\u8fa8\u8b58\u932f\u8aa4\u7684\u5f71\u97ff\u6bd4\u8f03\u5927\uff0c\u9020\u6210\u5176\u6458\u8981\u7d50\u679c\u4e0d\u5f70\u3002 \u7121\u8ad6\u5728 TD \u548c SD \u4e2d\uff0c\u4f7f\u7528 SMM \u8207 TriMM \u90fd\u6703\u6709\u6bd4 Top3 \u5dee\u7684\u6458\u8981\u6548\u80fd\uff0c\u56e0 \u70ba Gapped K \u662f\u4e00\u500b\u8f03\u4e0d\u7a69\u5b9a\u7684\u865b\u64ec\u76f8\u95dc\u6587\u4ef6\u9078\u53d6\u65b9\u6cd5\uff0c\u5728\u672c\u5be6\u9a57\u4e2d\u6709\u6bd4\u8f03\u5dee\u7684\u7d50\u679c\u662f \u53ef\u4ee5\u9810\u671f\u7684\u3002\u7fa4\u4e2d\u5fc3\u9078\u53d6\u6cd5(Centroid)\u8868\u73fe\u5c1a\u53ef\uff0c\u5728 TD \u53ca SD \u4e2d\uff0c\u65bc\u5404\u7a2e\u95dc\u806f\u6a21\u578b(RM\u3001",
                "content": "<table><tr><td/><td colspan=\"7\">\u8868\u4e00\u3001\u5be6\u9a57\u8a9e\u6599\u7d71\u8a08\u8cc7\u8a0a \u8868\u4e8c\u3001\u57fa\u790e\u5be6\u9a57\u7d50\u679c \u8868\u4e09\u3001\u95dc\u806f\u6a21\u578b\u4e4b\u5be6\u9a57\u7d50\u679c(\u4f7f\u7528\u6700\u9ad8\u6392\u5e8f\u6587\u4ef6\u524d\u4e09\u7bc7(Top3)) \u8868\u56db\u3001\u5404\u7a2e\u865b\u64ec\u76f8\u95dc\u6587\u4ef6\u9078\u53d6\u65b9\u6cd5\u65bc\u95dc\u806f\u6a21\u578b\u4e4b\u5be6\u9a57\u7d50\u679c</td></tr><tr><td colspan=\"2\">F-score</td><td/><td/><td/><td/><td>\u8a13\u7df4\u96c6</td><td>F-score F-score</td><td>\u6e2c\u8a66\u96c6</td></tr><tr><td/><td colspan=\"2\">\u8a9e\u6599\u6642\u9593</td><td colspan=\"5\">2001/11/07-2002/01/22 ROUGE-1 ROUGE-2 ROUGE-1 ROUGE-2</td><td>2002/01/23-2002/08/22 ROUGE-L ROUGE-L</td></tr><tr><td/><td colspan=\"2\">\u6587\u4ef6\u500b\u6578 LS KL</td><td/><td>0.411</td><td>0.225</td><td>185</td><td>0.098 0.298</td><td>20 0.183 0.361</td></tr><tr><td>TD</td><td colspan=\"2\">\u6587\u4ef6\u5e73\u5747\u6301\u7e8c\u5e7e\u79d2 \u6587\u4ef6\u5e73\u5747\u8a5e\u500b\u6578 LEAD RM</td><td/><td>0.450</td><td>0.310</td><td>129.4 326.0</td><td>0.194 0.336</td><td>141.2 290.3 0.276 0.400</td></tr><tr><td/><td colspan=\"2\">\u6587\u4ef6\u5e73\u5747\u8a9e\u53e5\u500b\u6578 VSM SMM</td><td/><td>0.436</td><td>0.347</td><td>20.0</td><td>0.228 0.325</td><td>23.3 0.290 0.385</td></tr><tr><td colspan=\"3\">\u6587\u4ef6\u5e73\u5747\u5b57\u932f\u8aa4\uf961 TD LSA TriMM</td><td/><td>0.457</td><td>0.362</td><td/><td>0.233 0.350</td><td>0.316 0.404</td></tr><tr><td colspan=\"4\">(Character Error Rate, CER) MMR KL</td><td>0.364</td><td>0.368</td><td>28.8%</td><td>0.248 0.210</td><td>29.8% 0.322 0.307</td></tr><tr><td colspan=\"3\">KL \u6587\u4ef6\u5e73\u5747\u8a5e\u932f\u8aa4\uf961 RM (Word Error Rate, WER) ILP SD SMM</td><td/><td>0.374 0.375</td><td>0.411 0.442</td><td>38.0%</td><td>0.298 0.226 0.337 0.221</td><td>0.361 0.321 39.4% 0.401 0.314</td></tr><tr><td/><td>TriMM</td><td>LS</td><td/><td>0.379</td><td>0.181</td><td/><td>0.044 0.228</td><td>0.138 0.325</td></tr><tr><td/><td colspan=\"2\">LEAD</td><td/><td/><td>0.255</td><td/><td>0.117</td><td>0.221</td></tr><tr><td/><td colspan=\"2\">VSM</td><td/><td/><td>0.342</td><td/><td>0.189</td><td>0.287</td></tr><tr><td colspan=\"8\">10%\uff0c\u5176\u5b9a\u7fa9\u70ba\u6458\u8981\u6240\u542b\u8a5e\u5f59\u6578\u5360\u6574 0.215 0.315 \u7bc7\u6587\u4ef6\u8a5e\u5f59\u6578\u7684\u6bd4\u4f8b\uff0c\u4e5f\u5c31\u662f\u4ee5\u8a5e\u5f59\u505a\u70ba\u5224\u65b7\u6458\u8981\u6bd4\u4f8b\u7684\u55ae\u5143\u3002\u5728\u6311\u9078\u6458\u8981\u8a9e\u53e5\u904e\u7a0b MMR 0.366 SD LSA 0.345 0.201 0.301 6.2\u3001\u57fa\u790e\u95dc\u806f\u6a21\u578b\u4e4b\u5be6\u9a57</td></tr><tr><td colspan=\"8\">\u4e2d\uff0c\u82e5\u9078\u5230\u67d0\u8a9e\u53e5\u4e2d\u7684\u67d0\u500b\u8a5e\u5f59\u6642\u5c31\u5df2\u7d93\u525b\u597d\u9054\u5230\u6458\u8981\u6bd4\u4f8b\uff0c\u70ba\u4e86\u4fdd\u6301\u8a9e\u53e5\u8a9e\u610f\u5b8c\u6574 KL 0.364 0.210 \u4f7f\u7528\u95dc\u806f\u6a21\u578b\u65bc\u8a9e\u53e5\u6a21\u578b\u4e4b\u5efa\u7acb\u6642\uff0c\u9700\u8981\u505a\u4e00\u6b21\u7684\u8cc7\u8a0a\u6aa2\u7d22\u4f86\u70ba\u6bcf\u500b\u8a9e\u53e5\u627e\u51fa\u865b\u64ec\u76f8\u95dc 0.307 \u6027\uff0c\u6b64\u8a9e\u53e5\u5269\u4e0b\u7684\u8a5e\u5f59\u4e5f\u6703\u88ab\u6311\u9078\u6210\u70ba\u6458\u8981\u3002 ILP 0.348 0.209 0.306 \u6587\u4ef6\uff0c\u7531\u540c\u6642\u671f\u7684\u65b0\u805e\u6587\u5b57\u6587\u4ef6(\u5171 101,268 \u7bc7)\u4e2d\u70ba\u6bcf\u4e00\u8a9e\u53e5\u9078\u53d6\u51fa 20 \u7bc7\u865b\u64ec\u76f8\u95dc\u6587</td></tr><tr><td colspan=\"8\">\u4ef6\uff0c\u4f46\u70ba\u4e86\u8981\u8207\u5f8c\u7e8c\u865b\u64ec\u76f8\u95dc\u6587\u4ef6\u9078\u53d6\u65b9\u6cd5\u4f5c\u516c\u5e73\u7684\u6bd4\u8f03\uff0c\u56e0\u6b64\u6b64\u57fa\u790e\u95dc\u806f\u6a21\u578b\u5be6\u9a57\u662f</td></tr><tr><td colspan=\"8\">\u53d6\u524d\u4e09\u7bc7(Top3)\u4f86\u9032\u884c\u95dc\u806f\u6a21\u578b\u4e4b\u4f30\u6e2c\u8207\u76f8\u95dc\u5be6\u9a57[4]\u3002\u7531\u65bc\u6587\u4ef6\u4e2d\u7684\u8a9e\u53e5\u901a\u5e38\u76f8\u5c0d\u7c21 \u516d\u3001\u5be6\u9a57\u7d50\u679c \u7684\u9650\u5236\u6027\u6700\u4f73\u5316(Constraint Optimization)\u7684\u8a9e\u53e5\u9078\u53d6\u65b9\u6cd5[22]\u3002 \u77ed\uff0c\u56e0\u6b64\u7576\u4f7f\u7528\u6700\u5927\u5316\u76f8\u4f3c\u5ea6\u4f30\u6e2c\u5efa\u7acb\u8a9e\u53e5\u6a21\u578b\u6642\uff0c\u5bb9\u6613\u906d\u9047\u8cc7\u6599\u7a00\u758f\u7684\u554f\u984c\uff0c\u4e0d\u5bb9\u6613 6.3\u3001\u5404\u7a2e\u865b\u64ec\u76f8\u95dc\u6587\u4ef6\u9078\u53d6\u65b9\u6cd5\u65bc\u95dc\u806f\u6a21\u578b\u4e4b\u5be6\u9a57</td></tr><tr><td colspan=\"8\">\u672c\u8ad6\u6587\u4e3b\u8981\u8457\u91cd\u5728\u865b\u64ec\u76f8\u95dc\u6587\u4ef6\u9078\u53d6\u65b9\u6cd5\u4e4b\u767c\u5c55\u8207\u6539\u9032\uff0c\u662f\u5c6c\u65bc\u975e\u76e3\u7763\u5f0f\u6458\u8981\u65b9\u6cd5\u7684\u7bc4 \u7587\uff0c\u56e0\u6b64\u6240\u6bd4\u8f03\u7684\u5c0d\u8c61\u4ee5\u975e\u76e3\u7763\u5f0f\u6458\u8981\u65b9\u6cd5\u70ba\u4e3b\uff1b\u9664\u6b64\u4e4b\u5916\uff0c\u672c\u8ad6\u6587\u4ea6\u5617\u8a66\u8207\u73fe\u4eca\u6700\u88ab \u5ee3\u70ba\u4f7f\u7528\u7684\u76e3\u7763\u5f0f\u6a5f\u5668\u5b78\u7fd2\u65b9\u6cd5\u505a\u6bd4\u8f03\uff0c\u5373\u652f\u6301\u5411\u91cf\u6a5f(SVM)[10]\u3002 6.1\u3001\u57fa\u790e\u5be6\u9a57 \u9996\u5148\uff0c\u6211\u5011\u6bd4\u8f03\u5eab\u723e\u8c9d\u514b-\u840a\u4f2f\u52d2\u96e2\u6563\u5ea6(KL)\u8207\u6578\u500b\u975e\u76e3\u7763\u5f0f\u6458\u8981\u65b9\u6cd5\u4e4b\u6458\u8981\u6210\u6548\uff0c\u5305 \u542b\u6709\u6700\u9577\u8a9e\u53e5\u6458\u8981(Longest Sentence, LS)\u3001\u9996\u53e5\u6458\u8981(LEAD)[26]\u3001\u5411\u91cf\u7a7a\u9593\u6a21\u578b(Vector Space Model, VSM)[8]\u3001\u6f5b\u85cf\u8a9e\u610f\u5206\u6790(Latent Semantic Analysis, LSA)[8]\u3001\u6700\u5927\u908a\u969b\u95dc\u806f (Maximal Marginal Relevance, MMR)[2]\u4ee5\u53ca\u6574\u6578\u7dda\u6027\u898f\u5283(Integer Linear Programming, ILP)[22]\u3002\u4e00\u822c\u4f86\u8aaa\uff0c\u6587\u4ef6\u4e2d\u9577\u53e5\u53ef\u80fd\u860a\u542b\u6709\u8f03\u8c50\u5bcc\u7684\u4e3b\u984c\u8cc7\u8a0a\uff0c\u56e0\u6b64\u4f9d\u64da\u6587\u4ef6\u4e2d\u8a9e\u53e5 \u9577\u5ea6\u505a\u6392\u5e8f\u5f8c\uff0c\u4f9d\u5e8f\u9078\u53d6\u6700\u9577\u8a9e\u53e5\u505a\u70ba\u6458\u8981\u7d50\u679c\u662f\u4e00\u7a2e\u7c21\u55ae\u7684\u6458\u8981\u65b9\u6cd5\u3002\u9664\u6b64\u4e4b\u5916\uff0c\u4e5f \u6709\u5b78\u8005\u7814\u7a76\u767c\u73fe\uff0c\u6587\u4ef6\u5e38\u4ee5\u958b\u9580\u898b\u5c71\u6cd5\u7684\u65b9\u5f0f\u4f86\u63d0\u9ede\u51fa\u4e3b\u984c\uff0c\u56e0\u6b64\u6587\u4ef6\u958b\u982d\u7684\u524d\u5e7e\u500b\u8a9e \u53e5\u7d93\u5e38\u662f\u5177\u4ee3\u8868\u6027\u7684\u8a9e\u53e5\uff0c\u9996\u53e5\u6458\u8981\u5373\u662f\u4ee5\u6b64\u6982\u5ff5\u51fa\u767c\uff0c\u9078\u53d6\u524d\u5e7e\u53e5\u8a9e\u53e5\u4f86\u5f62\u6210\u6574\u500b\u6587 \u4ef6\u7684\u6458\u8981\u3002\u6700\u9577\u8a9e\u53e5\u6458\u8981(LS)\u53ca\u9996\u53e5\u6458\u8981(LEAD)\u90fd\u50c5\u9069\u7528\u5728\u4e00\u90e8\u5206\u5177\u6709\u7279\u6b8a\u7d50\u69cb\u7684\u6587 \u4ef6\u4e0a\uff0c\u56e0\u6b64\u5b83\u5011\u7684\u7f3a\u9ede\u5c31\u662f\u6709\u5176\u4fb7\u9650\u6027\u3002\u53e6\u5916\uff0c\u5411\u91cf\u7a7a\u9593\u6a21\u578b\u662f\u628a\u6587\u4ef6\u548c\u8a9e\u53e5\u5206\u5225\u8996\u70ba \u901a\u5e38\u8a9e\u97f3\u6587\u4ef6\u4e3b\u8981\u6703\u6709\u8a9e\u97f3\u8fa8\u8b58\u932f\u8aa4\u548c\u8a9e\u53e5\u908a\u754c\u5075\u6e2c\u932f\u8aa4\u7684\u554f\u984c\uff0c\u4f46\u6211\u5011\u6709\u5148\u7d93\u4eba \u5de5\u5207\u97f3\uff0c\u56e0\u6b64\u6452\u9664\u4e86\u8a9e\u53e5\u908a\u754c\u5075\u6e2c\u932f\u8aa4\u7684\u554f\u984c\uff0c\u85c9\u7531\u6bd4\u8f03 TD \u8207 SD \u4e4b\u5be6\u9a57\u7d50\u679c\uff0c\u6211\u5011 \u7372\u5f97\u7cbe\u6e96\u7684\u6a21\u578b\uff0c\u6545\u6211\u5011\u671f\u671b\u8003\u616e\u984d\u5916\u7684\u95dc\u806f\u8cc7\u8a0a\u65bc\u8a9e\u97f3\u6587\u4ef6\u6458\u8981\uff0c\u4ea6\u5373\u85c9\u7531\u865b\u64ec\u76f8\u95dc \u672c\u5c0f\u7bc0\u7684\u5be6\u9a57\u662f\u7531\u7b2c\u4e00\u6b21\u865b\u64ec\u76f8\u95dc\u6587\u4ef6(\u6700\u9ad8\u6392\u5e8f\u6587\u4ef6) D Top \u4e2d(|D top |=20)\u518d\u7cbe\u934a\u9078\u53d6\u51fa \u6587\u4ef6\u4f86\u91cd\u65b0\u4f30\u6e2c\u4e26\u5efa\u7acb\u8a9e\u53e5\u7684\u8a9e\u8a00\u6a21\u578b\uff0c\u80fd\u7372\u5f97\u9032\u4e00\u6b65\u5730\u6458\u8981\u6210\u6548\u3002\u91cd\u65b0\u4f30\u6e2c\u5f8c\u7684\u95dc\u806f \u8f03\u4f73\u7684\u865b\u64ec\u76f8\u95dc\u6587\u4ef6 D P (|D P |=3)\uff0c\u6211\u5011\u6bd4\u8f03\u6240\u63d0\u51fa\u5169\u7a2e\u65b0\u7a4e\u7684\u9078\u53d6\u65b9\u6cd5(\u5373\u4e3b\u52d5\u5f0f-\u95dc\u806f \u6a21\u578b\u5247\u53ef\u8207\u539f\u672c\u7684\u8a9e\u53e5\u6a21\u578b\u76f8\u7d50\u5408\u6216\u53d6\u4ee3\u4e4b\uff0c\u76f8\u7d50\u5408\u7684\u53c3\u6578\u8abf\u6574\u5728\u672c\u5be6\u9a57\u4e2d\u662f\u63a1\u7528\u7d93\u9a57 \u591a\u5143\u5bc6\u5ea6\u975e\u76f8\u95dc(Active-RDDN)\u548c\u91cd\u758a\u5206\u7fa4(Overlapped))\u8207\u5176\u4ed6\u73fe\u6709\u9078\u53d6\u65b9\u6cd5(\u5373\u9593\u9694\u5f0f \u8a2d\u5b9a(Empirical Setting)\u3002\u5be6\u9a57\u7d50\u679c\u5982\u8868\u4e09\u6240\u793a\uff0c\u5728 TD \u8207 SD \u4e4b\u6458\u8981\u6210\u6548\u4e0a\uff0c\u4f7f\u7528\u95dc\u806f \u6700 \u9ad8 K \u9078 \u53d6 \u6cd5 (Gapped K) \u3001 \u7fa4 \u4e2d \u5fc3 \u9078 \u53d6 \u6cd5 (Centroid) \u4ee5 \u53ca \u4e3b \u52d5 \u5f0f -\u95dc \u806f \u591a \u5143 \u5bc6 \u5ea6 \u6a21\u578b(RM)\u3001\u7c21\u55ae\u6df7\u5408\u6a21\u578b(SMM)\u53ca\u4e09\u6df7\u5408\u6a21\u578b(TriMM)\u7686\u80fd\u6bd4\u57fa\u790e\u7684 KL \u5be6\u9a57\u8f03\u597d\uff0c\u5c24 (Active-RDD))\u65bc\u5404\u7a2e\u95dc\u806f\u6a21\u578b(RM\u3001SMM \u53ca TriMM)\u4e4b\u6458\u8981\u6548\u80fd\u6bd4\u8f03\uff0c\u5be6\u9a57\u7d50\u679c\u5982\u8868 \u5176\u662f\u4e09\u6df7\u5408\u6a21\u578b(TriMM)\u76f8\u8f03\u65bc KL \u5728 TD \u53ca SD \u7684 ROUGE-2 \u7d50\u679c\u4e0a\u80fd\u6709 5.2%\u8207 1.8% \u56db\u6240\u793a\uff0c\u8207\u57fa\u790e\u95dc\u806f\u6a21\u578b\u53ea\u4f7f\u7528\u524d\u4e09\u7bc7(Top3)\u865b\u64ec\u76f8\u95dc\u6587\u4ef6\u7684\u7d50\u679c\u76f8\u8f03(\u53c3\u7167\u8868\u4e09)\uff0c\u5927 \u7684\u6539\u9032\u3002\u63a5\u8457\uff0c\u6211\u5011\u6bd4\u8f03\u4e0d\u540c\u95dc\u806f\u6a21\u578b\u7684\u6458\u8981\u6210\u6548\uff0c\u9996\u5148\u662f\u95dc\u806f\u6a21\u578b(RM)\u8207\u7c21\u55ae\u6df7\u5408 \u90e8\u5206\u900f\u904e\u865b\u64ec\u76f8\u95dc\u6587\u4ef6\u9078\u53d6\u65b9\u6cd5\u90fd\u6703\u6bd4\u53ea\u4f7f\u7528 Top3 \u7684\u6458\u8981\u7d50\u679c\u9084\u8981\u4f86\u5f97\u597d\uff0c\u9664\u4e86 \u6a21\u578b(SMM)\u7684\u6bd4\u8f03\uff0c\u5f9e\u8868\u4e09\u7684\u5be6\u9a57\u7d50\u679c\u5f97\u77e5\u95dc\u806f\u6a21\u578b\u5728 TD \u4e0a\u8868\u73fe\u6bd4\u7c21\u55ae\u6df7\u5408\u6a21\u578b\u4f86\u5f97 \u597d\uff0c\u4f46\u5728 SD \u4f3c\u4e4e\u5728 ROUGE-1 \u5c31\u6c92\u6bd4\u7c21\u55ae\u6df7\u5408\u6a21\u578b\u597d\uff0c\u4e0d\u904e SD \u7684 ROUGE-2 \u8ddf ROUGE-L \u90fd\u9084\u662f\u6bd4\u7c21\u55ae\u6df7\u5408\u6a21\u578b\u7684\u6548\u679c\u597d\u3002\u95dc\u806f\u6a21\u578b\u7684\u5047\u8a2d\u662f\u5f37\u8abf\u8a5e\u5f59 w \u8207\u8a9e\u53e5 S \u5728 \u9019\u4e9b\u865b\u64ec\u76f8\u95dc\u6587\u4ef6\u4e2d\u540c\u6642\u51fa\u73fe\u4e4b\u95dc\u4fc2(\u53c3\u7167\u5f0f(10))\u4f86\u4f30\u6e2c\u6a21\u578b\uff0c\u800c\u7c21\u55ae\u6df7\u5408\u6a21\u578b\u662f\u5f37\u8abf Gapped K SMM \u53ca TriMM)\u4e0b\uff0c\u6bd4 Top3 \u53ca Gapped K \u90fd\u8981\u4f86\u5f97\u597d\u3002Active-RDD \u56e0\u5728\u9078\u53d6\u865b\u64ec\u76f8\u95dc \u8a13\u7df4\u597d\u7684\u6a21\u578b\u80fd\u8b93\u6709\u7368\u7279\u6027\u7684\u8a5e\u5f59\u5f97\u5230\u66f4\u591a\u7684\u6a5f\u7387\u503c\u56e0\u800c\u8b93\u6a21\u578b\u5177\u6709\u9451\u5225\u80fd\u529b\uff0c\u5169\u8005\u7686 \u6587\u4ef6\u6642\u540c\u6642\u8003\u91cf\u4e86\u95dc\u806f\u6027(Relevance)\u3001\u591a\u5143\u6027(Diversity)\u4ee5\u53ca\u5bc6\u5ea6\u6027(Density)\uff0c\u7528\u65bc\u4e0d\u540c \u6709\u5176\u597d\u8655\u3002\u6700\u5f8c\uff0c\u4e09\u6df7\u5408\u6a21\u578b(TriMM)\u56e0\u8907\u96dc\u5316\u4e86\u7c21\u55ae\u6df7\u5408\u6a21\u578b(SMM)\uff0c\u984d\u5916\u591a\u8003\u91cf\u6587 \u7684\u95dc\u806f\u6a21\u578b\u8a13\u7df4\u6642\uff0c\u76f8\u5c0d\u65bc Top3\u3001Gapped K \u4ee5\u53ca Centroid\uff0c\u7121\u8ad6\u5728 TD \u6216 SD \u4e2d\uff0c\u90fd\u53ef \u4ef6\u6a21\u578b\u7684\u5f71\u97ff\u529b\uff0c\u56e0\u6b64\u76f8\u8f03\u65bc\u95dc\u806f\u6a21\u578b\u53ca\u7c21\u55ae\u6df7\u5408\u6a21\u578b\u80fd\u5f97\u5230\u66f4\u4f73\u7684\u6458\u8981\u6548\u80fd\uff0c\u4e09\u6df7\u5408 \u4ee5\u5f97\u5230\u66f4\u597d\u7684\u6458\u8981\u7d50\u679c\u3002Active-RDDN \u5728\u591a\u8003\u91cf\u4e86\u975e\u76f8\u95dc(Non-relevance)\u8cc7\u8a0a\u7684\u60c5\u6cc1 \u6a21\u578b\u76f8\u8f03\u65bc\u95dc\u806f\u6a21\u578b\u5728 TD \u4e0a\u6709\u660e\u986f\u7684\u9032\u6b65\uff0c\u65bc ROUGE-2 \u7d50\u679c\u80fd\u6709 1.4%\u7684\u6539\u9032\uff0c\u4f46\u5728 \u4e0b\uff0c\u5176\u5be6\u9a57\u7d50\u679c\u90fd\u6703\u6bd4\u73fe\u6709\u7684\u9078\u53d6\u65b9\u6cd5\u8f03\u4f73\uff0c\u76f8\u5c0d\u65bc Active-RDD\u3001Centroid\u3001Gapped K SD \u4e0a\uff0c\u65bc ROUGE-2 \u7d50\u679c\u53ea\u6709\u5fae\u91cf\u7684 0.2%\u6539\u5584\u3002 \u4ee5\u53ca Top3 \u7121\u8ad6\u5728 TD \u6216 SD \u4e2d\uff0c\u5404\u7a2e\u95dc\u806f\u6a21\u578b(RM\u3001SMM \u53ca TriMM)\u4e0b\u90fd\u6703\u5f97\u5230\u6bd4\u8f03 \u4e00\u500b\u5411\u91cf\uff0c\u4e26\u4f7f\u7528\u8a5e\u983b-\u53cd\u6587\u4ef6\u983b(TF-IDF)\u7279\u5fb5\u4f86\u8a08\u7b97\u6bcf\u4e00\u7dad\u5ea6\u7684\u6b0a\u91cd\u503c\uff0c\u6587\u4ef6\u8207\u8a9e\u53e5 \u53ef\u4ee5\u89c0\u5bdf\u8a9e\u97f3\u8fa8\u8b58\u932f\u8aa4\u7387\u5c0d\u6458\u8981\u7d50\u679c\u7684\u5f71\u97ff\u6027\u3002\u6bd4\u8f03\u5404\u5f0f\u65b9\u6cd5\uff0cSD \u6bd4 TD \u4e0b\u964d\u4e86 \u5728\u95dc\u806f\u6a21\u578b\u7684\u76f8\u95dc\u5be6\u9a57\u4e2d\uff0c\u8a9e\u97f3\u8fa8\u8b58\u932f\u8aa4\u4e5f\u662f\u5f71\u97ff\u6458\u8981\u6548\u80fd\u975e\u5e38\u56b4\u91cd\uff0c\u5728\u4e09\u6df7\u5408\u6a21 \u597d\u7684\u6458\u8981\u7d50\u679c\uff0c\u6240\u4ee5\u8b49\u5be6\u975e\u76f8\u95dc\u8cc7\u8a0a\u78ba\u5be6\u4e00\u500b\u6709\u7528\u7684\u9078\u53d6\u7dda\u7d22\u3002\u6700\u5f8c\u672c\u8ad6\u6587\u6240\u63d0\u51fa\u7684\u91cd \u9593\u7684\u95dc\u806f\u6027\u662f\u85c9\u7531\u9918\u5f26\u76f8\u4f3c\u5ea6\u91cf\u503c\u4f86\u4f30\u6e2c\uff0c\u7576\u8a9e\u53e5\u5206\u6578\u8f03\u9ad8\u6642\uff0c\u5247\u8d8a\u6709\u6a5f\u6703\u6210\u70ba\u6b64\u6587\u4ef6 1.9%~8.8%\u7684 ROUGE-2 \u6458\u8981\u6548\u80fd\uff0c\u7531\u6b64\u53ef\u77e5\u8a9e\u97f3\u8fa8\u8b58\u932f\u8aa4\u7387\u5c0d\u6458\u8981\u6548\u80fd\u662f\u6709\u986f\u8457\u7684\u5f71 \u578b\u7684\u6578\u64da\u4e2d\uff0cSD \u6bd4 TD \u5287\u70c8\u4e0b\u964d\u4e86 12.2%\u7684 ROUGE-2 \u6458\u8981\u6548\u80fd\uff0c\u5728\u672a\u4f86\u7814\u7a76\u4e2d\uff0c\u6211\u5011 \u758a\u5206\u7fa4(Overlapped)\u9078\u53d6\u65b9\u6cd5\u7121\u8ad6\u5728 TD \u6216 SD \u4e2d\uff0c\u65bc\u5404\u7a2e\u95dc\u806f\u6a21\u578b\u4e0b(RM\u3001SMM \u53ca \u7684\u6458\u8981\u3002\u6f5b\u85cf\u8a9e\u610f\u5206\u6790\u662f\u5728\u5411\u91cf\u7a7a\u9593\u7684\u5047\u8a2d\u4e0b\u66f4\u9032\u4e00\u6b65\u5730\u4f7f\u7528\u5947\u7570\u503c\u5206\u89e3(Singular \u97ff\u6027\u3002\u70ba\u4e86\u6e1b\u7de9\u8a9e\u97f3\u8fa8\u8a8d\u932f\u8aa4\u7684\u554f\u984c\uff0c\u5728\u672a\u4f86\u6211\u5011\u5c07\u5617\u8a66\u4f7f\u7528\u97f3\u7bc0(Syllable)\u70ba\u55ae\u4f4d\u4f86\u5efa \u8a8d\u70ba\u53ef\u4ee5\u4ee5\u6b21\u8a5e\u7d22\u5f15(Subword Indexing)\u7684\u65b9\u5f0f\u4f86\u5efa\u7acb\u95dc\u806f\u6a21\u578b\u4ee5\u6e1b\u7de9\u8a9e\u97f3\u8fa8\u8b58\u932f\u8aa4\u4e4b TriMM)\u90fd\u53ef\u4ee5\u5f97\u5230\u6700\u597d\u7684\u6458\u8981\u6548\u679c\uff0c\u9a57\u8b49\u4e86\u91cd\u758a\u5206\u7fa4\u5728\u5229\u7528\u865b\u64ec\u76f8\u95dc\u6587\u4ef6\u4e2d\u7d50\u69cb\u5316\u8cc7 Value Decomposition, SVD)\u4f86\u627e\u5230\u53ef\u80fd\u7684\u6f5b\u85cf\u8a9e\u610f\u7a7a\u9593\uff0c\u4f7f\u4e4b\u80fd\u5728\u8003\u91cf\u6f5b\u85cf\u8a9e\u610f\u7684\u60c5\u6cc1 \u4e0b\u9032\u884c\u6587\u4ef6\u8207\u8a9e\u53e5\u7684\u95dc\u806f\u6027\u91cf\u6e2c\u3002\u6700\u5927\u908a\u969b\u95dc\u806f\u53ef\u8996\u70ba\u662f\u5411\u91cf\u7a7a\u9593\u6a21\u578b\u7684\u4e00\u500b\u5ef6\u4f38\uff0c\u5728 \u7acb\u8a9e\u53e5\u4ee5\u53ca\u6587\u4ef6\u6a21\u578b\uff1b\u6216\u5229\u7528\u8a5e\u5716(Word Graph)\u3001\u6df7\u6dc6\u7db2\u8def(Confusion Network)\u4f86\u542b\u62ec \u5f71\u97ff\u3002 \u8a0a\u78ba\u5be6\u53ef\u4ee5\u627e\u5230\u5177\u4ee3\u8868\u6027\u7684\u6587\u4ef6\u4ee5\u5229\u5404\u7a2e\u95dc\u806f\u6a21\u578b\u7684\u6a21\u578b\u8a13\u7df4\u6216\u53c3\u6578\u4f30\u6e2c\u3002</td></tr><tr><td colspan=\"8\">\u505a\u8a9e\u53e5\u6392\u5e8f\u6642\u8003\u91cf\u4e86\u5197\u9918\u6027\u4ee5\u9054\u5230\u66f4\u597d\u7684\u6458\u8981\u7d50\u679c\u3002\u6574\u6578\u7dda\u6027\u898f\u5283\u662f\u4e00\u500b\u5168\u57df(Global) \u66f4\u591a\u7684\u53ef\u80fd\u6b63\u78ba\u5019\u9078\u8a5e\u5f59\u4ee5\u88e8\u76ca\u6a21\u578b\u4f30\u6e2c\uff1b\u66f4\u53ef\u5229\u7528\u97fb\u5f8b\u8cc7\u8a0a(Prosodic Information)\u7b49</td></tr><tr><td colspan=\"8\">\u8072\u5b78\u7dda\u7d22\u4f86\u8f14\u52a9\u6e1b\u7de9\u8a9e\u97f3\u8fa8\u8a8d\u932f\u8aa4\u5c0d\u6458\u8981\u6548\u80fd\u7684\u5f71\u97ff\u3002</td></tr></table>",
                "html": null
            },
            "TABREF7": {
                "type_str": "table",
                "num": null,
                "text": "0.450 0.336 0.400 0.436 0.325 0.385 0.457 0.350 0.404",
                "content": "<table><tr><td colspan=\"2\">\u7684\u975e\u76e3\u7763\u5f0f\u6458\u8981\u65b9\u6cd5\u6709\u66f4\u52a0\u7684\u6458\u8981\u6548\u80fd\u8868\u73fe\u3002</td></tr><tr><td colspan=\"2\">\u672a\u4f86\uff0c\u6211\u5011\u7684\u7814\u7a76\u5c07\u6709\u4e09\u500b\u4e3b\u8981\u7684\u65b9\u5411\uff1a\u9996\u5148\uff0c\u672c\u8ad6\u6587\u6240\u63d0\u51fa\u4e4b\u865b\u64ec\u76f8\u95dc\u6587\u4ef6\u9078\u53d6</td></tr><tr><td colspan=\"2\">\u65b9\u6cd5\u662f\u5efa\u69cb\u5728\u5411\u91cf\u7a7a\u9593\u6216\u8a9e\u8a00\u6a21\u578b\u7a7a\u9593\u4e0a\uff0c\u4e26\u6c92\u6709\u8003\u616e\u5230\u8a9e\u610f\u7a7a\u9593\u7684\u76f8\u4f3c\u5ea6\u91cf\uff0c\u6211\u5011\u5c07</td></tr><tr><td colspan=\"2\">\u9032\u4e00\u6b65\u7684\u7814\u7a76\u662f\u5426\u53ef\u4ee5\u5728\u6f5b\u85cf\u8a9e\u610f\u7a7a\u9593\u4e2d\u4f86\u9078\u53d6\u8f03\u597d\u7684\u865b\u64ec\u76f8\u95dc\u6587\u4ef6\uff0c\u4ee5\u671f\u7372\u5f97\u66f4\u597d\u7684</td></tr><tr><td colspan=\"2\">\u6458\u8981\u6210\u6548\uff1b\u5176\u6b21\uff0c\u76ee\u524d\u6240\u767c\u5c55\u7684\u95dc\u806f\u6a21\u578b\u50c5\u904b\u7528\u65bc\u91cd\u5efa\u8a9e\u53e5\u7684\u8a9e\u8a00\u6a21\u578b\uff0c\u6211\u5011\u5c07\u5617\u8a66\u4f7f</td></tr><tr><td colspan=\"2\">Gapped K \u7528\u88ab\u6458\u8981\u6587\u4ef6\u7684\u95dc\u806f\u8cc7\u8a0a\u4f86\u91cd\u65b0\u4f30\u6e2c\u4e26\u5efa\u7acb\u6587\u4ef6\u7684\u8a9e\u8a00\u6a21\u578b\uff1b\u6700\u5f8c\uff0c\u6211\u5011\u5e0c\u671b\u80fd\u5c07\u975e\u76e3 0.451 0.338 0.401 0.433 0.317 0.385 0.454 0.343 0.406</td></tr><tr><td colspan=\"2\">Centroid \u7763\u5f0f\u65b9\u6cd5\u6240\u5f62\u6210\u7684\u7279\u5fb5\u7d50\u5408\u65bc\u66f4\u52a0\u8907\u96dc\u4e14\u6709\u6548\u7684\u76e3\u7763\u5f0f\u6a5f\u5668\u5b78\u7fd2\u65b9\u6cd5(\u5982 CRF \u6216\u6df1\u5ea6\u985e 0.449 0.334 0.402 0.439 0.331 0.389 0.456 0.353 0.407</td></tr><tr><td colspan=\"2\">Active-RDD 0.460 0.341 0.408 0.449 0.342 0.400 0.463 0.355 0.414 \u795e\u7d93\u7db2\u7d61(Deep Neural Network Learning, DNN)\u7b49)\u4e2d\uff0c\u4e26\u878d\u5408\u5176\u5b83\u8a9e\u97f3\u6587\u4ef6\u6240\u7368\u6709\u4e4b\u7279</td></tr><tr><td colspan=\"2\">Active-RDDN 0.464 0.352 0.411 0.455 0.346 0.405 0.466 0.367 0.421 \u5fb5(\u8af8\u5982\u97f3\u97fb\u8207\u8a9e\u8005\u7279\u5fb5\u7b49)\uff0c\u671f\u671b\u8a13\u7df4\u5f8c\u7684\u6a21\u578b\u80fd\u5920\u5728\u8a9e\u97f3\u6587\u4ef6\u6458\u8981\u4e0a\u7372\u5f97\u66f4\u597d\u7684\u8868</td></tr><tr><td>\u73fe\u3002</td><td/></tr><tr><td colspan=\"2\">Overlapped 0.470 0.354 0.416 0.460 0.341 0.410 0.471 0.362 0.422</td></tr><tr><td>Top3</td><td>0.374 0.226 0.321 0.375 0.221 0.314 0.379 0.228 0.325</td></tr><tr><td>Gapped K</td><td>0.374 0.228 0.322 0.371 0.218 0.313 0.376 0.225 0.315 \u5716\u4e8c\u3001SVM \u8207\u5176\u4ed6\u975e\u76e3\u7763\u5f0f\u6458\u8981\u65b9\u6cd5\u4e4b\u6bd4\u8f03</td></tr><tr><td colspan=\"2\">Centroid Active-RDD 0.379 0.228 0.332 0.378 0.229 0.321 0.388 0.242 0.335 0.374 0.227 0.314 0.377 0.227 0.320 0.380 0.233 0.328 6.4\u3001\u8207\u76e3\u7763\u5f0f\u6a21\u578b\u4e4b\u6bd4\u8f03 SD</td></tr><tr><td colspan=\"2\">Active-RDDN 0.383 0.239 0.330 0.380 0.226 0.327 0.391 0.244 0.339 \u9664\u4e86\u5404\u5f0f\u975e\u76e3\u7763\u5f0f\u6458\u8981\u65b9\u6cd5\u5916\uff0c\u672c\u8ad6\u6587\u4ea6\u5617\u8a66\u6bd4\u8f03\u652f\u6301\u5411\u91cf\u6a5f(SVM)\u65bc\u6587\u4ef6\u6458\u8981\u4e4b\u6210</td></tr><tr><td colspan=\"2\">Overlapped 0.386 0.239 0.334 0.382 0.236 0.332 0.396 0.250 0.345 \u6548 \uff0c \u6bd4 \u8f03 \u7684 \u5c0d \u8c61 \u6709 \u57fa \u790e KL \u4ee5 \u53ca \u4f7f \u7528 \u91cd \u758a \u5206 \u7fa4 \u9078 \u53d6 \u65b9 \u6cd5 \u65bc \u4e0d \u540c \u7684 \u95dc \u806f \u6a21 \u578b \u4e2d</td></tr><tr><td colspan=\"2\">(RM-Overlap\u3001SMM-Overlap \u548c TriMM-Overlap)\u3002\u652f\u6301\u5411\u91cf\u6a5f\u662f\u73fe\u4eca\u5e38\u898b\u7684\u76e3\u7763\u5f0f\u6a5f\u5668</td></tr><tr><td colspan=\"2\">\u5b78\u7fd2\u65b9\u6cd5\u4e4b\u4e00\uff0c\u8fd1\u5e74\u4f86\u5df2\u6709\u5b78\u8005\u5c07\u5176\u904b\u7528\u81f3\u6587\u4ef6\u6458\u8981\u9818\u57df\u4e4b\u4e2d[10]\u3002\u672c\u8ad6\u6587\u4f7f\u7528\u8a13\u7df4\u96c6</td></tr><tr><td colspan=\"2\">\u7684 185 \u7bc7\u6587\u4ef6\u9032\u884c\u652f\u6301\u5411\u91cf\u6a5f\u6a21\u578b\u7684\u8a13\u7df4\u8a9e\u6599\uff0c\u6211\u5011\u70ba\u6587\u4ef6\u4e2d\u7684\u6bcf\u4e00\u8a9e\u53e5\u62bd\u53d6 35 \u7dad\u7279</td></tr><tr><td colspan=\"2\">\u5fb5[13]\uff0c\u5305\u62ec\u6709\u97fb\u5f8b\u7279\u5fb5(Prosodic Features)\u3001\u8a9e\u5f59\u7279\u5fb5(Lexical Features)\u3001\u7d50\u69cb\u7279\u5fb5</td></tr><tr><td colspan=\"2\">(Structural Features)\u4ee5\u53ca\u57fa\u672c\u7684\u6a21\u578b\u7279\u5fb5(Model Features)\u7b49\u8cc7\u8a0a\uff0c\u5176\u6838\u5fc3\u51fd\u6578\u8a2d\u5b9a\u70ba\u534a</td></tr><tr><td colspan=\"2\">\u5f91\u5f0f\u51fd\u6578(Radial Basis Function)\uff0c\u5176\u4e2d SVM \u7684\u53c3\u6578\u8a2d\u5b9a\u90fd\u662f\u4f7f\u7528\u9810\u8a2d\u503c\u3002</td></tr><tr><td colspan=\"2\">\u5be6\u9a57\u7d50\u679c\u5982\u5716\u4e8c\u6240\u793a\u3002\u4e00\u5982\u9810\u671f\u5730\uff0cSVM \u8207\u5176\u4ed6\u5404\u5f0f\u975e\u76e3\u7763\u5f0f\u6a21\u578b\u76f8\u6bd4\u8f03\uff0c\u5728 TD</td></tr><tr><td colspan=\"2\">\u5be6\u9a57\u4e0a(\u5176 ROUGE-2 \u70ba 0.383)\u662f\u8868\u73fe\u6700\u597d\u7684\u65b9\u6cd5\uff0c\u9019\u662f\u7531\u65bc\u76e3\u7763\u5f0f\u6a5f\u5668\u5b78\u7fd2\u85c9\u7531\u4f7f\u7528\u4eba</td></tr><tr><td colspan=\"2\">\u5de5\u6a19\u6ce8\u7684\u6458\u8981\u53e5\u5b50\u9032\u884c\u6a21\u578b\u4e4b\u8a13\u7df4\uff0c\u5176\u4f7f\u7528\u7684\u8cc7\u8a0a\u8f03\u975e\u76e3\u7763\u5f0f\u6a5f\u5668\u5b78\u7fd2\u65b9\u6cd5\u591a\u4e14\u6b63\u78ba\uff0c</td></tr><tr><td colspan=\"2\">\u56e0\u6b64\u5176\u6458\u8981\u7684\u6548\u679c\u4e5f\u8f03\u975e\u76e3\u7763\u5f0f\u6a5f\u5668\u5b78\u7fd2\u4f86\u7684\u597d\u3002\u503c\u5f97\u4e00\u63d0\u7684\u662f\uff0c\u4f7f\u7528\u91cd\u758a\u5206\u7fa4\u865b\u64ec\u76f8</td></tr><tr><td colspan=\"2\">\u95dc\u6587\u4ef6\u9078\u53d6\u65b9\u6cd5\u65bc\u4e09\u6df7\u5408\u6a21\u578b\u4e2d(TriMM-Overlap)\uff0c\u6458\u8981\u4e4b\u6210\u6548\u5728 SD \u4e0a\u53ef\u6bd4\u76e3\u7763\u5f0f\u6a5f\u5668</td></tr><tr><td colspan=\"2\">\u5b78\u7fd2\u65b9\u6cd5\u7684 SVM \u4f86\u7684\u597d\u4e00\u4e9b\uff0c\u6b64\u4e00\u5be6\u9a57\u7d50\u679c\u4ee4\u4eba\u611f\u5230\u9a5a\u8a1d\uff0c\u56e0\u70ba\u672c\u8ad6\u6587\u6240\u63a2\u8a0e\u4e4b\u5404\u5f0f</td></tr><tr><td colspan=\"2\">\u6458\u8981\u65b9\u6cd5\u50c5\u8003\u616e\u4e86\u6587\u4ef6\u8207\u8a9e\u53e5\u4e2d\u7684\u55ae\u4e00\u7a2e\u7279\u5fb5\u503c\uff0c\u5373\u85c9\u7531\u8a5e\u5f59\u5206\u4f48\u8cc7\u8a0a\u4f86\u6311\u9078\u8a9e\u53e5\uff0c\u800c</td></tr><tr><td colspan=\"2\">\u652f\u6301\u5411\u91cf\u6a5f\u4e0d\u50c5\u4f7f\u7528\u4e86 35 \u7a2e\u7279\u5fb5\u503c\uff0c\u66f4\u9700\u8981\u4f7f\u7528\u4eba\u5de5\u6a19\u8a3b\u7684\u6b63\u78ba\u7b54\u6848\u9032\u884c\u6a21\u578b\u7684\u8a13</td></tr><tr><td colspan=\"2\">\u7df4\u3002\u6211\u5011\u8a8d\u70ba\uff0c\u6b64\u7d50\u679c\u4e4b\u539f\u56e0\u53ef\u80fd\u662f\u7531\u65bc\u652f\u6301\u5411\u91cf\u6a5f\u4e4b\u6458\u8981\u6280\u8853\u5728\u8a9e\u97f3\u8fa8\u8b58\u932f\u8aa4\u7684\u60c5\u6cc1</td></tr><tr><td colspan=\"2\">\u4e0b(\u5728\u6b64\u5be6\u9a57\u4e2d\uff0c\u8a13\u7df4\u96c6\u8207\u6e2c\u8a66\u96c6\u7684\u8a5e\u932f\u8aa4\u7387\u9054 40%) \uff0c\u672a\u5fc5\u80fd\u771f\u7684\u6709\u6548\u5b78\u7fd2\u5206\u8fa8\u6458\u8981</td></tr><tr><td colspan=\"2\">\u8207\u975e\u6458\u8981\u8a9e\u53e5\u3002</td></tr><tr><td colspan=\"2\">\u4e03\u3001\u7d50\u8ad6\u8207\u672a\u4f86\u65b9\u5411</td></tr><tr><td colspan=\"2\">\u672c\u8ad6\u6587\u57fa\u65bc\u8a9e\u8a00\u6a21\u578b\u5316\u67b6\u69cb\u4f86\u767c\u5c55\u8a9e\u97f3\u6458\u8981\u65b9\u6cd5\uff0c\u5176\u8ca2\u737b\u4e3b\u8981\u6709\u4e09\u65b9\u9762\u3002\u7b2c\u4e00\uff0c\u6709\u5225\u65bc</td></tr><tr><td colspan=\"2\">\u73fe\u6709\u57fa\u65bc\u8a9e\u8a00\u6a21\u578b\u5316\u67b6\u69cb\u4e4b\u6458\u8981\u65b9\u6cd5\u90fd\u805a\u7126\u5728\u8a9e\u53e5\u6a21\u578b\u53c3\u6578\u7684\u91cd\u65b0\u4f30\u6e2c\uff0c\u672c\u8ad6\u6587\u9996\u6b21\u6df1</td></tr><tr><td colspan=\"2\">\u5165\u63a2\u8a0e\u8207\u61c9\u7528\u5404\u7a2e\u65b0\u7a4e\u865b\u64ec\u76f8\u95dc\u6587\u4ef6\u9078\u53d6\u6280\u8853\u65bc\u7bc0\u9304\u5f0f\u8a9e\u97f3\u6587\u4ef6\u6458\u8981\u4efb\u52d9\uff0c\u7528\u4ee5\u5f37\u5316\u8a9e</td></tr><tr><td colspan=\"2\">\u53e5 \u6a21 \u578b \u7684 \u53c3 \u6578 \u4f30 \u6e2c \u3002 \u7b2c \u4e8c \uff0c \u672c \u8ad6 \u6587 \u66f4 \u9032 \u4e00 \u6b65 \u5730 \u8003 \u91cf \u4f7f \u7528 \u6bcf \u4e00 \u8a9e \u53e5 \u7684 \u975e \u76f8 \u95dc \u6027</td></tr><tr><td colspan=\"2\">(Non-relevance)\u8cc7\u8a0a\u5c0d\u65bc\u865b\u64ec\u76f8\u95dc\u6587\u4ef6\u9078\u53d6\u7684\u5f71\u97ff\u3002\u540c\u6642\uff0c\u6211\u5011\u4ea6\u984d\u5916\u5617\u8a66\u57fa\u65bc\u91cd\u758a\u5206</td></tr><tr><td colspan=\"2\">\u7fa4(Overlapped Clustering)\u6982\u5ff5\u4f86\u6709\u6548\u5730\u9078\u53d6\u91cd\u8981\u7684\u865b\u64ec\u76f8\u95dc\u6587\u4ef6\u3002\u7b2c\u4e09\uff0c\u672c\u8ad6\u6587\u63a2\u7d22\u4f7f</td></tr><tr><td colspan=\"2\">\u7528\u4e09\u6df7\u5408\u6a21\u578b(Tri-Mixture Model)\u4f86\u8868\u793a\u6bcf\u4e00\u8a9e\u53e5\uff0c\u671f\u76fc\u5176\u80fd\u66f4\u7cbe\u78ba\u5730\u8868\u793a\u8a9e\u53e5\u4e4b\u8a5e\u5f59</td></tr><tr><td colspan=\"2\">\u4f7f\u7528\u548c\u8a9e\u610f\u76f8\u95dc\u8cc7\u8a0a\u3002\u4e00\u7cfb\u5217\u7684\u5be6\u9a57\u7d50\u679c\u986f\u793a\uff0c\u672c\u8ad6\u6587\u6240\u63d0\u51fa\u4e4b\u65b9\u6cd5\u7684\u78ba\u80fd\u8f03\u5176\u5b83\u73fe\u6709</td></tr></table>",
                "html": null
            }
        }
    }
}