erhwenkuo commited on
Commit
4ff4991
1 Parent(s): 68bcf74

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +63 -1
README.md CHANGED
@@ -22,7 +22,69 @@ configs:
22
  data_files:
23
  - split: train
24
  path: data/train-*
 
 
 
 
 
 
 
25
  ---
26
  # Dataset Card for "c4-chinese-zhtw"
27
 
28
- [More Information needed](https://github.com/huggingface/datasets/blob/main/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
22
  data_files:
23
  - split: train
24
  path: data/train-*
25
+ task_categories:
26
+ - text-generation
27
+ - fill-mask
28
+ language:
29
+ - zh
30
+ size_categories:
31
+ - 1M<n<10M
32
  ---
33
  # Dataset Card for "c4-chinese-zhtw"
34
 
35
+ ## 內容
36
+
37
+ Common Crawl 是一個非營利組織,負責抓取網路並向公眾免費提供其檔案和資料集。Common Crawl 的網路檔案包含自 2008 年以來收集的 PB 級資料。它一般每月完成一次抓取。
38
+
39
+ Common Crawl 的爬蟲程式遵守 nofollow 和 robots.txt 政策。用於處理 Common Crawl 資料集的開源程式碼是公開可用的。
40
+
41
+ 這個繁中的數據來是來自 [Common Crawl](https://commoncrawl.org/overview) **2023-14** 的 data archive 下載并進行清理 。
42
+
43
+ 這是 [jed351](https://huggingface.co/jed351) 準備的版本,託管在這個位址:https://huggingface.co/datasets/jed351/Traditional-Chinese-Common-Crawl-Filtered
44
+
45
+ ## 支援的任務
46
+
47
+ C4主要用於預訓練語言模型(pretrain language model)。
48
+
49
+ ## 範例
50
+
51
+ 一個樣本的範例:
52
+
53
+ ```
54
+ {
55
+ 'url': 'http://www.bilingtong.com/cpzx/96.html',
56
+ 'timestamp': datetime.datetime(2023, 3, 21, 2, 12, 48),
57
+ 'content_language': 'zho',
58
+ 'content_type': 'text/plain',
59
+ 'text': '新風系統是通過系統設計送風和排風使室內空氣存在一空氣 。無需開窗全天持續不斷有組.....'
60
+ }
61
+ ```
62
+
63
+ ## 資料欄位
64
+
65
+ 資料有幾個欄位:
66
+
67
+ - `url`: 來源 url
68
+ - `timestamp`: 時間戳
69
+ - `content_language`: 內容包含的語言種類
70
+ - `content_type`: 內容類型,也稱為 MIME 或媒體類型,是 Web 伺服器回應標頭中的聲明
71
+ - `text`:網頁清理後的文字內容
72
+
73
+
74
+ ## 數據清理
75
+
76
+ 請參考在 Github 上的專案 [c4-dataset-script](https://github.com/jedcheng/c4-dataset-script) 來了解數據下載與清理的相關邏輯與程式碼。
77
+
78
+ 主要的步驟有:
79
+
80
+ 1. Download the WET crawl archive index file
81
+ 2. Run download and Chinese screening script on Spark
82
+ 3. Filter out non-sentence lines and toxic document
83
+ 4. Remove duplicated text
84
+ 5. Remove documents that are over self-repeating - Repetition Removal in DeepMind MassiveText
85
+
86
+ ## 許可資訊
87
+
88
+ 請尊循 Common Craw terms of use 的條款。
89
+
90
+ - https://commoncrawl.org/terms-of-use