Base_JSON/prefixO/json/O03/O03-1012.json

{
    "paper_id": "O03-1012",
    "header": {
        "generated_with": "S2ORC 1.0.0",
        "date_generated": "2023-01-19T08:01:03.792177Z"
    },
    "title": "Mencius: A Chinese Named Entity Recognizer Using Hybrid Model",
    "authors": [
        {
            "first": "Tzong-Han",
            "middle": [],
            "last": "Tsai",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "National Taiwan University Taipei",
                "location": {
                    "country": "Taiwan, R.O.C"
                }
            },
            "email": "thtsai@iis.sinica.edu.tw"
        },
        {
            "first": "Shih-Hung",
            "middle": [],
            "last": "Wu",
            "suffix": "",
            "affiliation": {},
            "email": "shwu@iis.sinica.edu.tw"
        },
        {
            "first": "Wen-Lian",
            "middle": [],
            "last": "Hsu",
            "suffix": "",
            "affiliation": {},
            "email": "hsu@iis.sinica.edu.tw"
        }
    ],
    "year": "",
    "venue": null,
    "identifiers": {},
    "abstract": "This paper presents a maximum entropy based Chinese named entity recognizer (NER): Mencius. It aims to address Chinese NER problems by combining the advantages of rule-based and machine learning (ML) based NER systems. Rule-based NER systems can explicitly encode human comprehension and can be tuned conveniently, while ML-based systems are robust, portable and inexpensive to develop. Our hybrid system incorporates a rule-based knowledge representation and template-matching tool, InfoMap [1], into a maximum entropy (ME) framework. Named entities are represented in InfoMap as templates, which serve as ME features in Mencius. These features are edited manually and their weights are estimated by the ME framework according to the training data. To avoid the errors caused by word segmentation, we model the NER problem as a character-based tagging problem. In our experiments, Mencius outperforms both pure rule-based and pure ME-based NER systems. The F-Measures of person names (PER), location names (LOC) and organization names (ORG) in the experiment are respectively 92.4%, 73.7% and 75.3%.",
    "pdf_parse": {
        "paper_id": "O03-1012",
        "_pdf_hash": "",
        "abstract": [
            {
                "text": "This paper presents a maximum entropy based Chinese named entity recognizer (NER): Mencius. It aims to address Chinese NER problems by combining the advantages of rule-based and machine learning (ML) based NER systems. Rule-based NER systems can explicitly encode human comprehension and can be tuned conveniently, while ML-based systems are robust, portable and inexpensive to develop. Our hybrid system incorporates a rule-based knowledge representation and template-matching tool, InfoMap [1], into a maximum entropy (ME) framework. Named entities are represented in InfoMap as templates, which serve as ME features in Mencius. These features are edited manually and their weights are estimated by the ME framework according to the training data. To avoid the errors caused by word segmentation, we model the NER problem as a character-based tagging problem. In our experiments, Mencius outperforms both pure rule-based and pure ME-based NER systems. The F-Measures of person names (PER), location names (LOC) and organization names (ORG) in the experiment are respectively 92.4%, 73.7% and 75.3%.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Abstract",
                "sec_num": null
            }
        ],
        "body_text": [
            {
                "text": "Information Extraction (IE) is the task of extracting information of interest from unconstrained text. IE involves two main tasks: the recognition of named entities, and the recognition of the relationships among these named entities. Named Entity Recognition (NER) involves the identification of proper names in text and their classification into different types of named entities (e.g., persons, organizations, locations). NER is not only important in IE [3] but also in lexical acquisition for the development of robust NLP systems [4] . Moreover, NER has proven fruitful for tasks such as documents indexing, and maintenance of databases containing identified named entities.",
                "cite_spans": [
                    {
                        "start": 457,
                        "end": 460,
                        "text": "[3]",
                        "ref_id": "BIBREF2"
                    },
                    {
                        "start": 535,
                        "end": 538,
                        "text": "[4]",
                        "ref_id": "BIBREF3"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "During the last decade, NER has drawn much attention at Message Understanding Conferences (MUC) [5] [6] . Both rule-based and machine learning NER systems have had some success. Previous rule-based approaches have used manually constructed finite state patterns, which match text against a sequence of words. Such system (like University of Edinburgh's LTG [7] ) do not need too much training data and can encode expert human knowledge. However, rule-based approaches lack robustness and portability. Each new source of text requires a significant tweaking of the rules to maintain optimal performance; the maintenance costs can be quite steep.",
                "cite_spans": [
                    {
                        "start": 96,
                        "end": 99,
                        "text": "[5]",
                        "ref_id": "BIBREF4"
                    },
                    {
                        "start": 100,
                        "end": 103,
                        "text": "[6]",
                        "ref_id": "BIBREF5"
                    },
                    {
                        "start": 357,
                        "end": 360,
                        "text": "[7]",
                        "ref_id": "BIBREF6"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "Another popular approach in NER is machine-learning (ML). ML is more attractive in that it is more portable and less expensive to maintain. The representative ML approaches used in NER are HMM (BBN's IdentiFinder in [8, 9] and Maximum Entropy (ME) (New York Univ.'s MEME in [10] [11] ). Although ML systems are relatively inexpensive to develop, the outputs of these systems are difficult to interpret.",
                "cite_spans": [
                    {
                        "start": 216,
                        "end": 219,
                        "text": "[8,",
                        "ref_id": null
                    },
                    {
                        "start": 220,
                        "end": 222,
                        "text": "9]",
                        "ref_id": "BIBREF9"
                    },
                    {
                        "start": 274,
                        "end": 278,
                        "text": "[10]",
                        "ref_id": "BIBREF10"
                    },
                    {
                        "start": 279,
                        "end": 283,
                        "text": "[11]",
                        "ref_id": "BIBREF12"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "As well, it is difficult to improve the system performance through error analysis. The performance of a ML system can be very poor when training data is insufficient.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "Furthermore, the performance of ML systems is worse than that of rule-based ones by about 2% as witnessed in MUC-6 [12] and MUC-7 [13] . This might be due to the fact that current ML approaches can capture non-parametric factors less effectively than human experts who handcraft the rules. Nonetheless, ML approaches do provide important statistical information that is unattainable by human experts. Currently, the F-measure in English rule-based and ML NER systems are 85% ~ 94% on MUC-7 data [14] . This is higher than the average performance of Chinese NER systems, which ranges from 79% to 86% [14] .",
                "cite_spans": [
                    {
                        "start": 115,
                        "end": 119,
                        "text": "[12]",
                        "ref_id": "BIBREF13"
                    },
                    {
                        "start": 130,
                        "end": 134,
                        "text": "[13]",
                        "ref_id": "BIBREF14"
                    },
                    {
                        "start": 495,
                        "end": 499,
                        "text": "[14]",
                        "ref_id": "BIBREF15"
                    },
                    {
                        "start": 599,
                        "end": 603,
                        "text": "[14]",
                        "ref_id": "BIBREF15"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "In this paper, we address the problem of Chinese NER. In Chinese sentences, there are no spaces between words, no capital letters to denote proper names or sentence breaks, and, worst of all, no standard definition of \"words\". As a result, word boundaries cannot, at times, be discerned without context. As well, the length of a named entity is longer on average than an English one, thus, the complexity of a Chinese NER system is greater.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "Previous works [15] [16] [2] on Chinese NER rely on the word segmentation module.",
                "cite_spans": [
                    {
                        "start": 15,
                        "end": 19,
                        "text": "[15]",
                        "ref_id": "BIBREF16"
                    },
                    {
                        "start": 20,
                        "end": 24,
                        "text": "[16]",
                        "ref_id": "BIBREF17"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "However, an error in the word segmentation step could lead to errors in NER results.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "Therefore, we bypass word segmentation and use a character-based tagger, treat each character as a token, and combine the tagged outcomes of continuing characters to form an NER output.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "Borthwick [11] uses an ME framework to integrate many NLP resources, including previous systems such as Proteus, a POS tagger. In this paper, Mencius incorporates a rule-based knowledge representation and template-matching tool, InfoMap [1] , into a maximum entropy (ME) framework. Named entities are represented in InfoMap as templates, which serve as ME features in Mencius. These features are edited manually and their weights are estimated by the ME framework according to the training data.",
                "cite_spans": [
                    {
                        "start": 10,
                        "end": 14,
                        "text": "[11]",
                        "ref_id": "BIBREF12"
                    },
                    {
                        "start": 237,
                        "end": 240,
                        "text": "[1]",
                        "ref_id": "BIBREF0"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "This paper is organized as follows. Section 2 provides the ME-based framework for NER. Section 3 describes features and how to represent them in our knowledge representation system, InfoMap. The data set and experimental results are discussed in Section 4. Section 5 gives our conclusions and possible extensions of the current work.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "For our purpose, we regard each character as a token. Consider a test corpus and a set of n named entity categories. Since a named entity can have more than one token, we associate two tags to each category x: x_begin and x_continue. In addition, we use the tag unknown to indicate that a token is not part of a named entity. The NER problem can then be rephrased as the problem of assigning one of 2n + 1 tags to each token. In Given a set of features and a training corpus, the ME estimation process produces a model in which every feature f i has a weight \u03b1 i . This allows us to compute the conditional probability as follows [17] .",
                "cite_spans": [
                    {
                        "start": 630,
                        "end": 634,
                        "text": "[17]",
                        "ref_id": "BIBREF18"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Maximum Entropy-Based NER Framework",
                "sec_num": "2."
            },
            {
                "text": "\u220f = i o h f i i h Z h o p ) , ( ) ( 1 ) | ( \u03b1 (2)",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Maximum Entropy-Based NER Framework",
                "sec_num": "2."
            },
            {
                "text": "Intuitively, the probability is the multiplication of weights of active features (i.e. those f i (h,o) = 1). The weight \u03b1 i is estimated by a procedure called Generalized Iterative Scaling (GIS) [18] . This is an iterative method that improves the estimation of the weights at each iteration. The ME estimation technique guarantees that for every feature f i , the expected value of\u03b1 i equals the empirical expectation of\u03b1 i in the training corpus.",
                "cite_spans": [
                    {
                        "start": 195,
                        "end": 199,
                        "text": "[18]",
                        "ref_id": "BIBREF19"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Maximum Entropy-Based NER Framework",
                "sec_num": "2."
            },
            {
                "text": "As Borthwick [11] remarked, ME allows the modeler to concentrate on finding the features that characterize the problem while letting the ME estimation routine deal with assigning relative weights to the features.",
                "cite_spans": [
                    {
                        "start": 13,
                        "end": 17,
                        "text": "[11]",
                        "ref_id": "BIBREF12"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Maximum Entropy-Based NER Framework",
                "sec_num": "2."
            },
            {
                "text": "After having trained an ME model and assigned the proper weight \u03b1 i to each feature f i , decoding (i.e. marking up) a new piece of text becomes a simple task. First, Mencius tokenizes the text and preprocesses the testing sentence. Then for each token we check which features are active and combine the\u03b1 i of the active features according to equation 2. Finally, a Viterbi search is run to find the highest probability path through the lattice of conditional probabilities that does not produce any invalid tag sequences (for instance the sequence [person_begin, location_continue] is invalid). Further details on the Viterbi search can be found in [19] .",
                "cite_spans": [
                    {
                        "start": 650,
                        "end": 654,
                        "text": "[19]",
                        "ref_id": "BIBREF20"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Decoding",
                "sec_num": "2.2"
            },
            {
                "text": "We divide features that can be used to recognize named entities into four categories according to whether they are external and whether they are category dependent.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Features",
                "sec_num": "3"
            },
            {
                "text": "McDonald defined internal and external features in [20] . The internal evidence is found within the entity, while the external evidence is gathered from its context. We use category-independent features to distinguish named entities from non-named entities (e.g., first-character-of-a-sentence, capital-letter, out-of-vocabulary), and category-dependent features to distinguish between different named entity categories (for example, surname and given name lists are used for recognizing person names).",
                "cite_spans": [
                    {
                        "start": 51,
                        "end": 55,
                        "text": "[20]",
                        "ref_id": "BIBREF21"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Features",
                "sec_num": "3"
            },
            {
                "text": "However, to simplify our design, we only use internal features that are category-dependent in this paper.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Features",
                "sec_num": "3"
            },
            {
                "text": "To calculate values of location features and organization features, Mencius uses InfoMap. InfoMap is our knowledge representation and template matching tool, which represents location or organization names as templates. An input string (sentence) is first matched to one or more location or organization templates by InfoMap and then passed to Mencius, there it is assigned feature values which further distinguish which named entity category it falls into.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "InfoMap -Our Knowledge Representation System",
                "sec_num": "3.1"
            },
            {
                "text": "InfoMap is a hierarchical knowledge representation scheme, consisting of several domains, each with a tree-like taxonomy. The basic units of information in InfoMap are called generic nodes which represent concepts, and function nodes which represent the relationships among generic nodes of one specific domain. In addition, generic nodes can also contain cross references to other nodes to avoid needless repetition.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Knowledge Representation Scheme in InfoMap",
                "sec_num": "3.1.1"
            },
            {
                "text": "In Mencius, we apply the geographical taxonomy of InfoMap called GeoMap. Our location and organization templates refer to generic nodes in Geomap. In Figure 1 , path that refers to the generic node \"Counties\". The second element is a wildcard ($$) which must be 2 to 4 characters in length. The third element is a specified character \"\u5c40\" (Department). ",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 150,
                        "end": 158,
                        "text": "Figure 1",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Knowledge Representation Scheme in InfoMap",
                "sec_num": "3.1.1"
            },
            {
                "text": "GeoMap",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Knowledge Representation Scheme in InfoMap",
                "sec_num": "3.1.1"
            },
            {
                "text": "In general, locations are divided into four types: administrative division, public area (park, airport, or port), landmark (road, road section, cross section or address), and landform (mountain, river, sea, or ocean). An administrative division name usually contains one or more than one location names in hierarchical order, such as \u5b89\u5927\uf976\uf96d \u591a\uf9d4\u591a\u5e02 (Toronto, Ontario while an organization keyword, \u516c\u53f8 (Corporation), is the right boundary marker.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Features for Recognizing Location Names",
                "sec_num": "3.2.2"
            },
            {
                "text": "The organization name in this category is led by one or more than one geographical names but the organization keyword (e.g., \u516c\u53f8 (Corporation)) is omitted. For example, \u53f0\u7063\u6377\u5b89\u7279 (Giant Taiwan) only contains the left boundary \u53f0\u7063 (Taiwan).",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Type II: With left boundary markers:",
                "sec_num": null
            },
            {
                "text": "The organization name in this category is ended by an organization keyword. For example, \u6377\u5b89\u7279\u516c\u53f8 (Giant Corporation) only contains the right boundary \u516c\u53f8 (Corporation). , tags (from t -2 to t 2 ), and organization keywords, e.g., \u516c\u53f8 (Corporation), to find the most likely tag sequences and recognize them. Table 4 . These features are helpful in recognizing organization names. Secondly, our corpus is much larger than MET2. MET2 contains 174 Chinese PER, 750 LOC, and 377 ORG while our corpus contains 1,242 Chinese PER, 954 LOC, and 1,147 ORG in 10,000 sentences (about 126,872 Chinese characters). The statistics of our data is shown in Table 5 . ",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 303,
                        "end": 310,
                        "text": "Table 4",
                        "ref_id": "TABREF6"
                    },
                    {
                        "start": 637,
                        "end": 644,
                        "text": "Table 5",
                        "ref_id": "TABREF7"
                    }
                ],
                "eq_spans": [],
                "section": "Type III: With right boundary marker:",
                "sec_num": null
            },
            {
                "text": "To demonstrate that Mencius performs better than pure rule-based and ML systems, we conduct the following three experiments. We use a 4-fold cross validation to test our system.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Experimental Results",
                "sec_num": "4.2"
            },
            {
                "text": "In this experiment, we use a person name list and InfoMap templates to recognize all named entities. The number of lexicons in person name lists and gazetteers is 32000. As shown in Table 6 , the results indicate the F-Measures of PER, LOC and ORG are 83.6%, 71.2% and 76.8%, respectively. ",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 182,
                        "end": 189,
                        "text": "Table 6",
                        "ref_id": "TABREF8"
                    }
                ],
                "eq_spans": [],
                "section": "Name Lists and Templates (Rule-based)",
                "sec_num": "4.2.1"
            },
            {
                "text": "In this experiment, we apply the pure ME model, which only uses context information of characters from c -2 to c 2 and tags from t -2 to t 2 . As shown in ",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Pure Maximum Entropy Model (ML-based)",
                "sec_num": "4.2.2"
            },
            {
                "text": "In this section, we discuss problems encountered by Mencius.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Discussions",
                "sec_num": "4.3"
            },
            {
                "text": "As shown in Tables 6, 7 ",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 12,
                        "end": 23,
                        "text": "Tables 6, 7",
                        "ref_id": "TABREF8"
                    }
                ],
                "eq_spans": [],
                "section": "Data Sparseness",
                "sec_num": "4.3.1"
            },
            {
                "text": "In this section, we show error cases associated with each named entity category.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Other Errors",
                "sec_num": "4.3.2"
            },
            {
                "text": "The summary report in Table 8 shows that the precision and recall rates for person names are 97.9% and 87.4%, respectively. The major errors are listed below.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 22,
                        "end": 29,
                        "text": "Table 8",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "A. Person Names",
                "sec_num": null
            },
            {
                "text": "(1) The surname character of a person name is not in surname list or the given-name character is not in the given-name character list. Therefore, some of the person features are not set to 1. For example, \uf9e1\u54d6 (Lee Nian) are not recognized because \u54d6 (Nian) is not in the given-name character list. ",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "A. Person Names",
                "sec_num": null
            },
            {
                "text": "The summary report in Table 8 shows the precision and recall rates for location names are 78.6% and 69.4%, respectively. The major errors are listed below.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 22,
                        "end": 29,
                        "text": "Table 8",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "B. Location Names",
                "sec_num": null
            },
            {
                "text": "(1) Location names within an organization name are extracted but the organization name is not recognized. For example, \u97d3\u570b\u6771\u6d0b\u88fd\u679c (Korea Orion Food) is not recognized as an organization name, but \u97d3\u570b (Korea) is recognized as a location name. (2) The location name is abbreviated. For example, \u53f0 (Tai), the abbreviated form of \u53f0 \u7063 (Taiwan), is not recognized in some cases.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "B. Location Names",
                "sec_num": null
            },
            {
                "text": "(3) The Chinese usually call a market street. For example, \u96fb\u5b50\u8857 (Electronics St.)",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "B. Location Names",
                "sec_num": null
            },
            {
                "text": "represents an electronics market. However, this is an informal name. Table 8 shows the precision and recall rate for organization name recognition are 94.4% and 62.6%, respectively. We illustrate standard error analysis with examples.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 69,
                        "end": 76,
                        "text": "Table 8",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "B. Location Names",
                "sec_num": null
            },
            {
                "text": "(1) The organization name is a bilingual term. For example, eBay \u53f0\u7063 (eBay Taiwan) is not recognized.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 74,
                        "end": 81,
                        "text": "Taiwan)",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "C. Organization Names",
                "sec_num": null
            },
            {
                "text": "(2) The organization name is in Type II, III, or IV category (defined in Section 3. ",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "C. Organization Names",
                "sec_num": null
            },
            {
                "text": "In this paper, we developed a Chinese NER system, Mencius, which does not rely on the word segmentation module. Instead, we model the NER problem as a character-based tagging problem. Mencius uses ME modeling combining advantages of rule-based and ML-based NER systems. Our hybrid system uses a rule-based knowledge representation system, InfoMap, and incorporates it into the ME framework.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Conclusions",
                "sec_num": "5"
            },
            {
                "text": "The F-Measures of person names (PER), location names (LOC) and organization names (ORG) in the experiment are respectively 92.4%, 73.7% and 75.3%. These are comparatively better than the results obtained by pure rule-based and pure ME-based method.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Conclusions",
                "sec_num": "5"
            }
        ],
        "back_matter": [
            {
                "text": "We are persuaded Mencius can be improved in the following directions. We only use internal features that are category-dependent in this version. In the future, we will collect more features, especially external ones. In addition, we will design a post-processing module to deal with the data sparseness problem. Moreover, we will use document level context information to recognize abbreviated names which cannot be recognized at present.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "acknowledgement",
                "sec_num": null
            }
        ],
        "bib_entries": {
            "BIBREF0": {
                "ref_id": "b0",
                "title": "FAQ-centered Organizational Memory",
                "authors": [
                    {
                        "first": "S",
                        "middle": [
                            "H"
                        ],
                        "last": "Wu",
                        "suffix": ""
                    },
                    {
                        "first": "M",
                        "middle": [
                            "Y"
                        ],
                        "last": "Day",
                        "suffix": ""
                    },
                    {
                        "first": "T",
                        "middle": [
                            "H"
                        ],
                        "last": "Tsai",
                        "suffix": ""
                    },
                    {
                        "first": "W",
                        "middle": [
                            "L"
                        ],
                        "last": "Hsu",
                        "suffix": ""
                    }
                ],
                "year": 2002,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "S. H. Wu, M. Y. Day, T. H. Tsai, and W. L. Hsu, \"FAQ-centered Organizational Memory,\" in Knowledge Management and Organizational Memories, R. Dieng-Kuntz, Ed. Boston: Kluwer Academic Publishers, 2002.",
                "links": null
            },
            "BIBREF1": {
                "ref_id": "b1",
                "title": "Chinese Named Entity Identification Using Class-based Language Model",
                "authors": [
                    {
                        "first": "J",
                        "middle": [],
                        "last": "Sun",
                        "suffix": ""
                    },
                    {
                        "first": "J",
                        "middle": [
                            "F"
                        ],
                        "last": "Gao",
                        "suffix": ""
                    },
                    {
                        "first": "L",
                        "middle": [],
                        "last": "Zhang",
                        "suffix": ""
                    },
                    {
                        "first": "M",
                        "middle": [],
                        "last": "Zhou",
                        "suffix": ""
                    },
                    {
                        "first": "C",
                        "middle": [
                            "N"
                        ],
                        "last": "Huang",
                        "suffix": ""
                    }
                ],
                "year": 2002,
                "venue": "presented at the 19th International Conference on Computational Linguistics",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "J. Sun, J. F. Gao, L. Zhang, M. Zhou, and C. N. Huang, \"Chinese Named Entity Identification Using Class-based Language Model,\" presented at the 19th International Conference on Computational Linguistics,, 2002.",
                "links": null
            },
            "BIBREF2": {
                "ref_id": "b2",
                "title": "in Information Extraction: A Multidisciplinary Approach to an Emerging Information Technology",
                "authors": [
                    {
                        "first": "R",
                        "middle": [],
                        "last": "Grishman",
                        "suffix": ""
                    }
                ],
                "year": 1997,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "10--26",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "R. Grishman, \"Information Extraction: Techniques and Challenges,\" in Information Extraction: A Multidisciplinary Approach to an Emerging Information Technology, J. G. Carbonell, Ed. Frascati, Italy: Springer, 1997, pp. 10-26.",
                "links": null
            },
            "BIBREF3": {
                "ref_id": "b3",
                "title": "The Analysis and Acquisition of Proper Names for Robust Text Understanding",
                "authors": [
                    {
                        "first": "S",
                        "middle": [],
                        "last": "Coates-Stephens",
                        "suffix": ""
                    }
                ],
                "year": 1992,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "S. Coates-Stephens, \"The Analysis and Acquisition of Proper Names for Robust Text Understanding,\" in Dept. of Computer Science. London: City University, 1992.",
                "links": null
            },
            "BIBREF4": {
                "ref_id": "b4",
                "title": "MUC-6 Named Entity Task Definition (Version 2.1),\" presented at the 6th Message Understanding Conference",
                "authors": [
                    {
                        "first": "N",
                        "middle": [],
                        "last": "Chinchor",
                        "suffix": ""
                    }
                ],
                "year": 1995,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "N. Chinchor, \"MUC-6 Named Entity Task Definition (Version 2.1),\" presented at the 6th Message Understanding Conference, Columbia, Maryland, 1995.",
                "links": null
            },
            "BIBREF5": {
                "ref_id": "b5",
                "title": "MUC-7 Named Entity Task Definition (Version 3.5),\" presented at the 7th Message Understanding Conference",
                "authors": [
                    {
                        "first": "N",
                        "middle": [],
                        "last": "Chinchor",
                        "suffix": ""
                    }
                ],
                "year": 1998,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "N. Chinchor, \"MUC-7 Named Entity Task Definition (Version 3.5),\" presented at the 7th Message Understanding Conference, Fairfax, Virginia, 1998.",
                "links": null
            },
            "BIBREF6": {
                "ref_id": "b6",
                "title": "Description of the LTG System Used for MUC-7",
                "authors": [
                    {
                        "first": "A",
                        "middle": [],
                        "last": "Mikheev",
                        "suffix": ""
                    },
                    {
                        "first": "C",
                        "middle": [],
                        "last": "Grover",
                        "suffix": ""
                    },
                    {
                        "first": "M",
                        "middle": [],
                        "last": "Moensk",
                        "suffix": ""
                    }
                ],
                "year": 1998,
                "venue": "presented at the 7th Message Understanding Conference",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "A. Mikheev, C. Grover, and M. Moensk, \"Description of the LTG System Used for MUC-7,\" presented at the 7th Message Understanding Conference, Fairfax, Virginia, 1998.",
                "links": null
            },
            "BIBREF8": {
                "ref_id": "b8",
                "title": "BBN: Description of the SIFT System as Used for MUC-7",
                "authors": [
                    {
                        "first": "",
                        "middle": [],
                        "last": "Weischedel",
                        "suffix": ""
                    }
                ],
                "year": 1998,
                "venue": "presented at the 7th Message Understanding Conference",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Weischedel, \"BBN: Description of the SIFT System as Used for MUC-7,\" presented at the 7th Message Understanding Conference, Fairfax, Virginia, 1998.",
                "links": null
            },
            "BIBREF9": {
                "ref_id": "b9",
                "title": "An Algorithm that Learns What's in a Name",
                "authors": [
                    {
                        "first": "D",
                        "middle": [],
                        "last": "Bikel",
                        "suffix": ""
                    },
                    {
                        "first": "R",
                        "middle": [],
                        "last": "Schwartz",
                        "suffix": ""
                    },
                    {
                        "first": "R",
                        "middle": [],
                        "last": "Weischedel",
                        "suffix": ""
                    }
                ],
                "year": 1999,
                "venue": "Machine Learning",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "D. Bikel, R. Schwartz, and R. Weischedel, \"An Algorithm that Learns What's in a Name,\" Machine Learning, 1999.",
                "links": null
            },
            "BIBREF10": {
                "ref_id": "b10",
                "title": "NYU: Description of the MENE Named Entity System as Used in MUC-7",
                "authors": [
                    {
                        "first": "A",
                        "middle": [],
                        "last": "Borthwick",
                        "suffix": ""
                    },
                    {
                        "first": "J",
                        "middle": [],
                        "last": "Sterling",
                        "suffix": ""
                    },
                    {
                        "first": "E",
                        "middle": [],
                        "last": "Agichtein",
                        "suffix": ""
                    },
                    {
                        "first": "R",
                        "middle": [],
                        "last": "Grishman",
                        "suffix": ""
                    }
                ],
                "year": null,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "A. Borthwick, J. Sterling, E. Agichtein, and R. Grishman, \"NYU: Description of the MENE Named Entity System as Used in MUC-7,\" presented at the 7th",
                "links": null
            },
            "BIBREF11": {
                "ref_id": "b11",
                "title": "Message Understanding Conference",
                "authors": [],
                "year": 1998,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Message Understanding Conference, Fairfax, Virginia, 1998.",
                "links": null
            },
            "BIBREF12": {
                "ref_id": "b12",
                "title": "A Maximum Entropy Approach to Named Entity Recognition",
                "authors": [
                    {
                        "first": "A",
                        "middle": [],
                        "last": "Borthwick",
                        "suffix": ""
                    }
                ],
                "year": 1999,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "A. Borthwick, \"A Maximum Entropy Approach to Named Entity Recognition,\" New York University, 1999.",
                "links": null
            },
            "BIBREF13": {
                "ref_id": "b13",
                "title": "Statistical Significance of MUC-6 Results",
                "authors": [
                    {
                        "first": "N",
                        "middle": [],
                        "last": "Chinchor",
                        "suffix": ""
                    }
                ],
                "year": 1995,
                "venue": "presented at the 6th Message Understanding Conference",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "N. Chinchor, \"Statistical Significance of MUC-6 Results,\" presented at the 6th Message Understanding Conference, Columbia, Maryland, 1995.",
                "links": null
            },
            "BIBREF14": {
                "ref_id": "b14",
                "title": "Statistical Significance of MUC-7 Results",
                "authors": [
                    {
                        "first": "N",
                        "middle": [],
                        "last": "Chinchor",
                        "suffix": ""
                    }
                ],
                "year": 1998,
                "venue": "presented at the 7th Message Understanding Conference",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "N. Chinchor, \"Statistical Significance of MUC-7 Results,\" presented at the 7th Message Understanding Conference, Fairfax, Virginia, 1998.",
                "links": null
            },
            "BIBREF15": {
                "ref_id": "b15",
                "title": "MUC-7 Test Score Reports for all Participants and all Tasks",
                "authors": [
                    {
                        "first": "N",
                        "middle": [],
                        "last": "Chinchor",
                        "suffix": ""
                    }
                ],
                "year": 1998,
                "venue": "presented at the 7th Message Understanding Conference",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "N. Chinchor, \"MUC-7 Test Score Reports for all Participants and all Tasks,\" presented at the 7th Message Understanding Conference, Fairfax, Virginia, 1998.",
                "links": null
            },
            "BIBREF16": {
                "ref_id": "b16",
                "title": "Description of the NTU System Used for MET2",
                "authors": [
                    {
                        "first": "H",
                        "middle": [
                            "H"
                        ],
                        "last": "Chen",
                        "suffix": ""
                    },
                    {
                        "first": "Y",
                        "middle": [
                            "W"
                        ],
                        "last": "Ding",
                        "suffix": ""
                    },
                    {
                        "first": "S",
                        "middle": [
                            "C"
                        ],
                        "last": "Tsai",
                        "suffix": ""
                    },
                    {
                        "first": "G",
                        "middle": [
                            "W"
                        ],
                        "last": "Bian",
                        "suffix": ""
                    }
                ],
                "year": 1998,
                "venue": "presented at the 7th Message Understanding Conference",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "H. H. Chen, Y. W. Ding, S. C. Tsai, and G. W. Bian, \"Description of the NTU System Used for MET2,\" presented at the 7th Message Understanding Conference, Fairfax, Virginia, 1998.",
                "links": null
            },
            "BIBREF17": {
                "ref_id": "b17",
                "title": "Description of the Kent Ridge Digital Labs System Used for MUC-7",
                "authors": [
                    {
                        "first": "S",
                        "middle": [
                            "H"
                        ],
                        "last": "Yu",
                        "suffix": ""
                    },
                    {
                        "first": "S",
                        "middle": [
                            "H"
                        ],
                        "last": "Bai",
                        "suffix": ""
                    },
                    {
                        "first": "P",
                        "middle": [],
                        "last": "Wu",
                        "suffix": ""
                    }
                ],
                "year": 1998,
                "venue": "presented at the 7th Message Understanding Conference",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "S. H. Yu, S. H. Bai, and P. Wu, \"Description of the Kent Ridge Digital Labs System Used for MUC-7,\" presented at the 7th Message Understanding Conference, Fairfax, Virginia, 1998.",
                "links": null
            },
            "BIBREF18": {
                "ref_id": "b18",
                "title": "A maximum entropy approach to natural language processing",
                "authors": [
                    {
                        "first": "A",
                        "middle": [],
                        "last": "Berger",
                        "suffix": ""
                    },
                    {
                        "first": "S",
                        "middle": [
                            "A"
                        ],
                        "last": "Della Pietra",
                        "suffix": ""
                    },
                    {
                        "first": "V",
                        "middle": [
                            "J"
                        ],
                        "last": "Della Pietra",
                        "suffix": ""
                    }
                ],
                "year": 1996,
                "venue": "Computational Linguistics",
                "volume": "22",
                "issue": "",
                "pages": "39--71",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "A. Berger, S. A. Della Pietra, and V. J. Della Pietra, \"A maximum entropy approach to natural language processing,\" Computational Linguistics, vol. 22, pp. 39-71, 1996.",
                "links": null
            },
            "BIBREF19": {
                "ref_id": "b19",
                "title": "Generalized iterative scaling for log-linear models",
                "authors": [
                    {
                        "first": "J",
                        "middle": [
                            "N"
                        ],
                        "last": "Darroch",
                        "suffix": ""
                    },
                    {
                        "first": "D",
                        "middle": [],
                        "last": "Ratcliff",
                        "suffix": ""
                    }
                ],
                "year": 1972,
                "venue": "Annals of Mathematicl Statistics",
                "volume": "43",
                "issue": "",
                "pages": "1470--1480",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "J. N. Darroch and D. Ratcliff, \"Generalized iterative scaling for log-linear models,\" Annals of Mathematicl Statistics, vol. 43, pp. 1470-1480, 1972.",
                "links": null
            },
            "BIBREF20": {
                "ref_id": "b20",
                "title": "Error Bounds for Convolutional Codes and an Asymptotically Optimum Decoding Algorithm",
                "authors": [
                    {
                        "first": "A",
                        "middle": [
                            "J"
                        ],
                        "last": "Viterbi",
                        "suffix": ""
                    }
                ],
                "year": 1967,
                "venue": "IEEE Transactions on Information Theory",
                "volume": "IT",
                "issue": "",
                "pages": "260--269",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "A. J. Viterbi, \"Error Bounds for Convolutional Codes and an Asymptotically Optimum Decoding Algorithm,\" IEEE Transactions on Information Theory, vol. IT, pp. 260-269, 1967.",
                "links": null
            },
            "BIBREF21": {
                "ref_id": "b21",
                "title": "Internal and External Evidence in the Identification and Semantic Categorization of Proper Names",
                "authors": [
                    {
                        "first": "D",
                        "middle": [],
                        "last": "Mcdonald",
                        "suffix": ""
                    }
                ],
                "year": 1996,
                "venue": "Corpus Processing for Lexical Acquisition",
                "volume": "",
                "issue": "",
                "pages": "21--39",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "D. McDonald, \"Internal and External Evidence in the Identification and Semantic Categorization of Proper Names,\" in Corpus Processing for Lexical Acquisition, J. Pustejovsky, Ed. Cambridge, MA: MIT Press, 1996, pp. 21-39.",
                "links": null
            }
        },
        "ref_entries": {
            "FIGREF0": {
                "type_str": "figure",
                "uris": null,
                "num": null,
                "text": "-Person-Surname c 0 c 1 c 2 or c 0 c 1 are in the name list \"\u9673\"\u6c34\u6241, \"\uf99a\"\u6230 Probably the first character of a person name Current-Char-Person-Given-Name c -2 c -1 c 0 or c -1 c 0 or c -1 c 0 c 1 are in the name list \u9673\"\u6c34\"\u6241, \u9673\u6c34\"\u6241\", \uf99a\"\u6230\" Probably the second or third character of a person name Current-Char-Surname c Current-Char-Person-Surname: This feature is set to 1 if c 0 c 1 c 2 or c 0 c 1 are in the person name database. For example, in the case c 0 c 1 c 2 = \u9673 \u6c34 \u6241 , the feature Current-Char-Person-Surname for \u9673 is active since c 0 and its following characters c 1 c 2 satisfy the feature condition. Current-Char-Person-Given-Name: This feature is set to 1 if c -2 c -1 c 0 , c -1 c 0 , or c -1 c 0 c 1 are in the person name database. Current-Char-Surname: This feature is set to 1 if c 0 is in the top 300 popular surname list. Current-Char-Given-Name: This feature is set to 1 if c 0 c 1 or c -1 c 0 are in the given name database. Current-Char-Freq-Given-Name-Character: (c 0 and c 1 ) or (c -1 and c 0 ) are in the frequently given name character list Current-Char-Speaking-Verb: c 0 or c 0 c 1 or c -1 c 0 are in the speaking verb list. This feature distinguishes a trigram containing a speaking verb such as \u9673\u6c96\uf96f (Chen Chong said) from a real person name. Current-Char-Title: c 0 or c 0 c 1 or c -1 c 0 are in the title list. This feature distinguishes a trigram containing a title such as \u9673\u5148\u751f (Mr. Chen) from a real person name."
            },
            "FIGREF1": {
                "type_str": "figure",
                "uris": null,
                "num": null,
                "text": "-InfoMap-Organization-Start of the first character is set to 1. In addition, the feature Current-Character-InfoMap-Organization-Continue of the remaining characters is set to 1. The necessary conditions for each organization feature and examples of matched data are shown in"
            },
            "FIGREF2": {
                "type_str": "figure",
                "uris": null,
                "num": null,
                "text": "For Chinese NER, the most famous corpus is MET-2[6]. There are two main differences between our corpus and MET-2: the number of domains and the amount of data. First, MET-2 contains only one domain (Accident) while our corpus, which is collected from the online United Daily News in December 2002 (http://www.udn.com.tw), contains six domains: Local News, Social Affairs, Investment, Politics, Headline news and Business, which provides more varieties of organization names than single domain corpus does. The full location names and organization names are comparatively longer in length and our corpus contains more location names under county level and addresses. Therefore, the patterns of location names and organization names are more complex in our corpus."
            },
            "FIGREF3": {
                "type_str": "figure",
                "uris": null,
                "num": null,
                "text": "Mencius outperforms the rule-based method (Lists and Templates) and ML-based method (pure ME) in the total F-Measure. However, rule-based approach outperforms Mencius in the ORG category. It is due to the data sparseness problem. For example, \u4e2d\u58e2\u5929\u665f\u91ab\u9662 is tagged as [organization_begin, Because \u4e2d\u58e2\u5929\u665f\u91ab\u9662 rarely occurs, it might not appear as an organization name in training set during the 4-fold cross validation experiment. The Viterbi search cannot deal with sequences containing unknown tags. With an appropriate post-processing procedure, this kind of error can be resolved. We can treat the unknown tag as x_continue in a certain window size."
            },
            "FIGREF4": {
                "type_str": "figure",
                "uris": null,
                "num": null,
                "text": "A person name follows a single-character word which can be a surname. For example, \u6234 is both a surname (Dai) and a verb (wear) in Chinese lexicon.However, in \u982d\u6234\uf9e1\u61c9\u5143\u7684\u5e3d\u5b50 (wear Lee Ying Yuan's Hat), \u6234 means wear while Mencius mistakenly considers \u6234 as a surname. Therefore, Mencius mistakenly recognizes \u6234\uf9e1\u61c9 (Dai Lee-Ying) as a person name rather than the correct person name \uf9e1\u61c9 \u5143 (Lee, Ying-Yuan). Several person names appear consecutively while all of their given names are omitted. Since the context of two person names and one person name are similar, Mencius may mistakenly extract an incorrect name. For example, in the sentence \u5433\u3001\u9ec3\u4e8c\u4eba\u5728\u4ed6\u5c31\u8077\u524d\uf978\u5929 , Mencius extracts \u9ec3\u4e8c\u4eba from it. However, \u4e8c\u4eba in English means \"both\", not the given name. Transliterated names are not defined in the person name category in Mencius. However, some transliterated person names look like Chinese person names. Therefore, Mencius mistakenly extracts \u67ef\uf9f4\u9813 (Clinton), \u590f\u99a8 (Shaheen) from sentences. Some Japanese and Korean person names look like Chinese person names. For example, Mencius mistakenly extracts \uf933\u6b66\u9249 (Roh, Moo Hyun) from sentences."
            },
            "FIGREF5": {
                "type_str": "figure",
                "uris": null,
                "num": null,
                "text": "2.3). For example, \u97d3\u570b\u6771\u6d0b\u88fd\u679c (Korea Orion Food), \u6771\u6d0b\u88fd\u679c\u516c\u53f8 (Korea Orion Food Corporation), and \u6771\u6d0b\u88fd\u679c (Orion Food). (3) Several organization names appear consecutively while part of each name is omitted. For example, in \u53f0\uf963\u5e02\uf9c4\u5b89, \u4fe1\u7fa9, \u5433\u8208\u7b49\u570b\u5c0f (Taipei Long-Ann, Hsin-Yi, and Wu-Xin elementary schools) , \uf9c4\u5b89 (Long-Ann), \u4fe1\u7fa9 (Xin-Yi) and \u5433\u8208 (Wu-Xin) are not recognized as organizations because the organization ending boundary markers are abbreviated. The organization name is a foreign organization name, which is not considered by our organization template. For example, \u65e5\u672c\u8fb2\uf9f4\u4e2d\u592e\uf90a\u5eab (The Norinchukin Bank) is not recognized as an organization name. (5) The organization name is an exception. In \u53f0\uf963\u7e23\u7b2c\u4e8c\u6240\u570b\u4e2d (the second junior high school in Taipei county), \u7b2c\u4e8c\u6240 means \"the second\", and appear in the wildcard part of template [ \u901a \u7528 \u5730 \uf9e4 . \u53f0 \u7063 . \u7e23 ]:$$(2..13): \u570b \u4e2d ([GeoMap.Taiwan.Counties]: $$(2..13):Junior-High-School). We need more out of vocabulary (OOV) knowledge to represent all the number plus quantifier patterns."
            },
            "TABREF0": {
                "content": "<table><tr><td>Mencius, there are 3 named entity categories and 7 tags: person_begin, person_continue, location_begin, location_continue, organization_begin, organization_continue and unknown. For example, the phrase [\uf9e1 \u9060 \u54f2 \u5728 \u9ad8 \u96c4 \u5e02] (Lee, Yuan Tseh in Kaohsiung City) could be tagged as [person_begin, person_continue, person_continue, unknown, location_begin, location_continue, location_continue]. helpful in making a prediction about the outcome. For instance, one of our features is: when the current character is a known surname, it is likely to be the leading character of a person name. More formally, we can represent this feature as \u23a9 \u23a8 \u23a7 = = = else : 0 _ and true Surname(h) -Char -Current if : 1 ) , ( begin person o o h f (1) Here, Current-Char-Surname(h) is a binary function that returns the value true if the 2.1 The computation of p(o|h) in ME depends on a set of binary-valued features, which are current character of the history h is in the surname list.</td></tr></table>",
                "type_str": "table",
                "num": null,
                "text": "Maximum Entropy ME is a flexible statistical model which assigns an outcome for each token based on its history and features. Outcome space is comprised of the seven Mencius tags for an ME formulation of NER. ME computes the probability p(o|h) for any o from the space of all possible outcomes O, and for every h from the space of all possible histories H. A history is all the conditioning data that enables one to assign probabilities to the space of outcomes. In NER, history could be viewed as all information derivable from the test corpus relative to the current token.",
                "html": null
            },
            "TABREF1": {
                "content": "<table/>",
                "type_str": "table",
                "num": null,
                "text": "Taoyuan County, etc. In InfoMap, we refer to generic nodes (or concept node) by paths. A path of generic nodes consists of all node names from the root of the domain to the specific generic node, in which function nodes are omitted. The node names are separated by periods. For example, the path for the",
                "html": null
            },
            "TABREF2": {
                "content": "<table><tr><td colspan=\"2\">Symbol Semantics</td><td>Example Template</td><td>Sample Matching</td></tr><tr><td/><td/><td/><td>String</td></tr><tr><td>:</td><td>Concatenate two strings</td><td>A:B</td><td>AB</td></tr><tr><td colspan=\"2\">$$(m..n) Wildcards (number of</td><td>A:$$(1..2):B</td><td>ACB, ADDB,</td></tr><tr><td/><td>characters can be from m to n;</td><td/><td>ACDB</td></tr><tr><td/><td>both m and n have to be</td><td/><td/></tr><tr><td/><td>non-negative integers)</td><td/><td/></tr><tr><td>[p]</td><td>A path to a generic node.</td><td>[GeoMap.Taiwan.Counties]</td><td>Taipei County,</td></tr></table>",
                "type_str": "table",
                "num": null,
                "text": "InfoMap template syntax Mencius only deals with surname plus first name (usually with two characters), for example, \u9673\u6c34\u6241 (Chen Shui-bian). There are various way to express a person in a sentence, such as \u9673\u5148\u751f (Mr. Chen) and\uf934\u9673 (Old Chen), which have not been incorporated into the current system. Furthermore, we do not target transliterated names, such as \u5e03\u5e0c (Bush), since they do not follow Chinese name composition rules.We use a table of frequently occurring names to process our candidate test data. If a character and its context (history) correspond to a feature condition, the value of the current character for that feature will be set to 1. Feature conditions, examples and explanations for each feature are shown inTable 2. In the feature conditions column, c -1 , c 0 , and c 1 represent the preceding character, the current character, and the following character respectively.",
                "html": null
            },
            "TABREF3": {
                "content": "<table><tr><td>features associated with it. The first is Current-Char-InfoMap-x-Begin, which is set to 1 for the first character of matched string and set to 0 for the remaining characters. The other is Current-Char-InfoMap-x-Continue, which is set to 1 for all the characters of matched string except for the first character and set to 0 for the first character. The intuition is: using InfoMap to help ME detect which character in the sentence is the first character of a location name and which characters are the remaining characters of a \u23a9 \u23a8 \u23a7 = = = else : 0 _ and true Begin -x -InfoMap -Char -Current if : 1 ) , ( begin x o o h f (3) In For each InfoMap template category x (e.g., location and organization), there are two \u23a9 \u23a8 \u23a7 = = = else : 0 _ and true Continue -x -InfoMap -Char -Current if : 1 ) , ( continue x o o h f (4)</td></tr></table>",
                "type_str": "table",
                "num": null,
                "text": "Mencius, we build templates to recognize three types of location names. Our administrative division templates contain more than one set of location names in hierarchical order. For example, the template, [\u901a\u7528\u5730\uf9e4.\u53f0\u7063.\u5e02]:[ \u901a\u7528\u5730\uf9e4.\u53f0\u7063. \u7528 \u5730 \uf9e4 . \u53f0 \u7063 . \u516c \u5712 ] ([GeoMap.Taiwan.Cities]:[GeoMap.Taiwan.Parks]) is for recognizing all Taiwanese city parks. Landmark templates are built in the same way. E.g., [\u901a\u7528\u5730\uf9e4.\u53f0\u7063.\u5e02]:$$(2..4):\uf937 ([GeoMap.Taiwan.Cities]:$$(2..4):Road), is for recognizing roads in Taiwan. Continue of the remaining characters of the matched string is set to 1.Table 3shows the necessary conditions for each organization feature and gives examples of matched data.",
                "html": null
            },
            "TABREF4": {
                "content": "<table><tr><td>Feature</td><td>Feature Conditions</td><td>Example</td><td>Explanations</td></tr><tr><td>Current-Char-InfoMap-Location-Begin</td><td>c 0~cn-1 matches an</td><td>\"\u53f0\"\uf963\u7e23\u677f</td><td>Probably the</td></tr><tr><td/><td>InfoMap location</td><td>\u6a4b\u5e02</td><td>leading character</td></tr><tr><td/><td>template, where the</td><td/><td>of a location</td></tr><tr><td/><td>character length of</td><td/><td/></tr><tr><td/><td>the template is n</td><td/><td/></tr><tr><td colspan=\"2\">Current-Char-InfoMap-Location-Continue c a \u2026c 0 \u2026 . c b matches</td><td>\u53f0\"\uf963\"\u7e23\u677f</td><td>Probably the</td></tr><tr><td/><td>an InfoMap location</td><td>\u6a4b\u5e02</td><td>continuing</td></tr></table>",
                "type_str": "table",
                "num": null,
                "text": "Location Features Organizations include named corporate, governmental, or other organizational entity.The difficulty of recognizing an organization name is that an organization name is usually led by location names, such as \u53f0 \uf963 \u5e02 \u5730 \u6aa2 \u7f72 (Taipei District Public Prosecutors Office). Therefore, traditional machine learning NER systems only identify the location part rather than the full organization name. For example, the system only extracts \u53f0\uf963\u5e02 (Taipei City) from \u53f0\uf963\u5e02 SOGO \u767e\u8ca8\u9031\uf98e\u6176 (Taipei SOGO Department Store Anniversary) rather than \u53f0\uf963\u5e02 SOGO \u767e\u8ca8 (Taipei SOGO Department Store). According to our analysis of the structure of Chinese organization",
                "html": null
            },
            "TABREF5": {
                "content": "<table><tr><td>Type IV: No boundary marker:</td></tr></table>",
                "type_str": "table",
                "num": null,
                "text": "In this category, both left and right boundaries as above mentioned are omitted, such as \u6377\u5b89\u7279 (Giant). The organization names in this category are usually in the abbreviated form. In Mencius, we build templates for recognizing Type I organization names. Each organization template begins with a location name in GeoMap and ends with an organization keyword. For example, we build [ \u901a \u7528 \u5730 \uf9e4 . \u53f0 \u7063 . \u5e02 ]:$$(2..4): \u5c40 ([GeoMap.Taiwan.Cities]:$$(2..4):Department) for recognizing county level government departments in Taiwan. However, in Type II, III, IV, organization names cannot be recognized by templates. Therefore, the maximum entropy model uses features of characters (from c -2 to c 2)",
                "html": null
            },
            "TABREF6": {
                "content": "<table><tr><td>Feature</td><td colspan=\"2\">Feature Conditions Example</td><td>Explanations</td></tr><tr><td>Current-Char-InfoMap-Organization-Begin</td><td>c 0~cn-1 is matches</td><td>\"\u53f0\"\uf963\u5e02</td><td>Probably the</td></tr><tr><td/><td>an InfoMap</td><td>\u6377\u904b\u516c\u53f8</td><td>leading character</td></tr><tr><td/><td>organization</td><td/><td>of an</td></tr><tr><td/><td>template, where the</td><td/><td>organization</td></tr><tr><td/><td>character length of</td><td/><td/></tr><tr><td/><td>the template is n</td><td/><td/></tr><tr><td colspan=\"2\">Current-Char-InfoMap-Organization-Continue c a \u2026c 0 \u2026 . c b</td><td/><td/></tr><tr><td/><td>matches an</td><td/><td/></tr><tr><td/><td>InfoMap</td><td/><td/></tr><tr><td/><td>organization</td><td/><td/></tr><tr><td/><td>template, where a is</td><td/><td/></tr><tr><td/><td>a negative integer</td><td/><td/></tr><tr><td/><td>and b is a</td><td/><td/></tr><tr><td/><td>non-negative</td><td/><td/></tr><tr><td/><td>integer</td><td/><td/></tr></table>",
                "type_str": "table",
                "num": null,
                "text": "Organization Features",
                "html": null
            },
            "TABREF7": {
                "content": "<table><tr><td/><td/><td colspan=\"2\">Statistics of Data Set</td><td/></tr><tr><td>Domain</td><td colspan=\"2\">Number of Named Entities</td><td/><td>Size (in</td></tr><tr><td/><td>PER</td><td>LOC</td><td>ORG</td><td>characters)</td></tr><tr><td>Local News</td><td>84</td><td>139</td><td>97</td><td>11835</td></tr><tr><td>Social Affairs</td><td>310</td><td>287</td><td>354</td><td>37719</td></tr><tr><td>Investment</td><td>20</td><td>63</td><td>33</td><td>14397</td></tr><tr><td>Politics</td><td>419</td><td>209</td><td>233</td><td>17168</td></tr><tr><td colspan=\"2\">Headline News 267</td><td>70</td><td>243</td><td>19938</td></tr><tr><td>Business</td><td>142</td><td>186</td><td>187</td><td>25815</td></tr><tr><td>Total</td><td>1242</td><td>954</td><td>1147</td><td>126872</td></tr></table>",
                "type_str": "table",
                "num": null,
                "text": "",
                "html": null
            },
            "TABREF8": {
                "content": "<table><tr><td/><td colspan=\"3\">Performance of Name Lists and Templates</td></tr><tr><td>NE</td><td>P(%)</td><td>R(%)</td><td>F(%)</td></tr><tr><td>PER</td><td>72.98</td><td>97.93</td><td>83.63</td></tr><tr><td>LOC</td><td>67.96</td><td>74.67</td><td>71.16</td></tr><tr><td>ORG</td><td>95.77</td><td>64.07</td><td>76.78</td></tr><tr><td>Total</td><td>75.62</td><td>82.13</td><td>78.74</td></tr></table>",
                "type_str": "table",
                "num": null,
                "text": "",
                "html": null
            },
            "TABREF9": {
                "content": "<table><tr><td/><td colspan=\"3\">Table 7. Performance of Pure Maximum Entropy</td></tr><tr><td>NE</td><td>P(%)</td><td>R(%)</td><td>F(%)</td></tr><tr><td>PER</td><td>62.38</td><td>21.64</td><td>32.13</td></tr><tr><td>LOC</td><td>72.83</td><td>18.31</td><td>29.26</td></tr><tr><td>ORG</td><td>38.24</td><td>1.15</td><td>2.23</td></tr><tr><td>Total</td><td>65.03</td><td>13.89</td><td>22.89</td></tr><tr><td colspan=\"4\">4.2.3 Integrating Name Lists and Templates into A Maximum Entropy-Based</td></tr><tr><td colspan=\"2\">Framework (Hybrid)</td><td/><td/></tr><tr><td colspan=\"4\">In this experiment, we integrate name lists, location templates, and organization</td></tr><tr><td colspan=\"4\">templates into a maximum-Entropy-Based framework. As shown in Table 8, the results</td></tr><tr><td colspan=\"4\">indicate that the performance of PER, LOC, ORG is better than those in 4.2.1 and 4.2.2.</td></tr><tr><td/><td/><td>Table 8. Hybrid Performance</td><td/></tr><tr><td>NE</td><td>P(%)</td><td>R(%)</td><td>F(%)</td></tr><tr><td>PER</td><td>97.94</td><td>87.39</td><td>92.36</td></tr><tr><td>LOC</td><td>78.60</td><td>69.35</td><td>73.69</td></tr><tr><td>ORG</td><td>94.39</td><td>62.57</td><td>75.25</td></tr><tr><td>Total</td><td>90.56</td><td>73.70</td><td>81.26</td></tr></table>",
                "type_str": "table",
                "num": null,
                "text": ", the results indicate that the F-Measures of PER, LOC and ORG are 32.1%, 29.3% and 2.2%, respectively.",
                "html": null
            }
        }
    }
}