學術產出-學位論文
文章檢視/開啟
書目匯出
-
題名 基於兩詞彙的序列關係建造非監督式 SeqWORDS 斷詞方法
SeqWORDS: an unsupervised Chinese segmentation method using relationship of two consecutive words.作者 吳冠輝
Wu, Guan-Hui貢獻者 薛慧敏
Hsueh, Huey-Miin
吳冠輝
Wu, Guan-Hui關鍵詞 中文斷詞
文本探勘
動態規劃法
文字詞典模型
EM演算法
詞彙序列關係
Chinese texts mining
Dynamic programming
EM algorithm
Word dictionary model
Words dependency
Word segmentation日期 2019 上傳時間 1-七月-2019 10:43:44 (UTC+8) 摘要 由於中文文本中的詞彙之間沒有任何標記或空格，所以斷詞被認為是中文文本探勘前必要且重要的預處理步驟。而目前中文斷詞方法多屬監督式方法，當沒有適當的詞典時難以發揮，例如針對新世代的文章或特定專業領域的文本。Deng等人在2016年提出非監督式斷詞方法TopWORDS，利用文字詞典模型(Word D ictionary Model, WDM)建構文本之概似函數，並且將斷詞資訊當作遺失變數，以EM演算法估計出各詞彙的使用機率，更利用動態規劃法(dynamic programm ing)計算，除了運算上相當具有效率，TopWORDS應用在許多文本上有良好的結果。然而，TopWORDS假設文本中每個位置的詞彙獨立且分配相同，這樣的假設恐怕忽略了詞彙在文意上的相連。此研究假設每個詞彙出現的概率與前一個詞彙有關，因此文本的概似函數可表示為兩詞彙的序列關係的函數，故將此研究提出的方法稱為「SeqWORDS」。在運用三種不同斷詞法於紅樓夢文本上後，我們觀察到 SeqWORDS雖然在探索新詞彙的能力較弱，然而當接續使用文本探勘工具如詞向量分析後發現，SeqWORDS 能提供最佳的解釋性。
Unlike alphabet-based language, there exists no space between words inChinese corpus. The first step in Chinese text mining is to segment words in a sentence. Many existing segmentation methods are supervised in terms of requiring an adequate dictionary. However, Chinese language has developed so long and growing so fast. A suitable dictionary may not be available or easily accessed. In 2016, Deng et al. proposed an unsupervised method called “TopWORDS”, which needs no dictionary in hand. The authors derived the likelihood function of the corpus via word dictionary model (WDM). Further, they regard unknown segmentation information as missing data and utilize EM algorithm to estimate occurrence probability of words. To enhance computational efficiency, the estimates are computed by dynamic programming. In the article, the TopWORDS is found to perform well in several corpus. However, the iid assumption of TopWORDS ignores words dependency, which frequently occurs in consecutive words. Therefore, in this research we assume that a word’s occurrence depends on previous one and modify the TopWORDS method. By considering the sequential association of consecutive words, the proposed method is named “SeqWORDS”. The new method and two other existing methods are evaluated by their performance on the famous classical novel Story-of-Stone. We find that SeqWORDS is less capable to find new, rare words and is much time consuming. However, when we further implement some advance text mining analysis on the segmented corpus, the segmented corpus by SeqWORDS produces the most reasonable, interpretable results.參考文獻 [1]The Stanford Natural Language. Processing Group, Chinese Natural Language Processing and Speech Processing. Retrieved May 24, 2019, from https://nlp.stanford.edu/projects/chinese-nlp.shtml#cws[2]J. Lafferty, A. McCallum, F. C.N. Pereira, (2001), Conditional random fields: Probabilistc models for segmenting and labeling sequence data. Proceedings of the 18th International Conference on Machine Learning 2001(ICML 2001), pp 282–289.[3]fxsjy, Jieba, Retrieved May 27, 2019, from https://github.com/fxsjy/ji eba[4]L. R. Rabiner, B. H. Juang, (1986), An introduction to hidden Markov models, IEEE ASSP MAGAZINE, vol 3, no 1, pp. 4-16.[5]A. Chen, (2003), Chinese word segmentation using minimal linguistic knowledge. Proceeding SIGHAN `03 Proceedings of the second SIGHAN workshop on Chinese language processing, Vol 17, pp 148–151.[6]K. J. Chen, S. H. Liu, (1992), Word identification for Mandarin Chinese sentences. Proceeding COLING `92 Proceedings of the 14th conference on Computational linguistics, Vol 1, pp 101–107.[7]K. Deng, P. K. Bol, K. J. Li, and J. S. Liu, (2016). On the unsupervised analysis of domain-specific Chinese texts. Proceedings of the National Academy of Sciences of the United States of America, vol 113, pp 6154–6159.[8]X. Ge, W. Pratt, P. Smyth, (1999), Discovering Chinese words from unsegmented text. Proceeding SIGIR `99 Proceedings of the 22nd annual international ACM SIGIR conference on Research and development in information retrieval, pp 271–272.[9]A. P. Dempster, N. M. Laird, D. B. Rubin, (1977), Maximum likelihood from incomplete data via the EM algorithm. Journal of the Royal Statistical Society. Series B, vol 39, no 1, pp 1-38.[10]R. Bellman, (1954), The theory of dynamic programming, Bulletin of the American Mathematical Society, vol 60, no 6, pp 503-515.[11]X. Cao, Story-of-Stone.[12] 胡適，(1988)，胡適紅樓夢研究論述全編，上海古籍出版社。[13]T. Mikolov, K. Chen, G. Corrado, J. Dean, (2013). Efficient Estimation of Word Representations in Vector Space, arXiv:1301.3781v3.[14]T. Mikolov, I. Sutskever, K. Chen, G. Corrado, J. Dean, (2013). DistributedRepresentations of Words and Phrases and their Compositionality, NIPS 2013,3111-3119.[15]K. Pearson, (1901), On Lines and Planes of Closest Fit to Systems of Points in Space. Philosophical Magazine, vol 2, pp 559-572. 描述 碩士
國立政治大學
統計學系
106354027資料來源 http://thesis.lib.nccu.edu.tw/record/#G0106354027 資料類型 thesis dc.contributor.advisor 薛慧敏 zh_TW dc.contributor.advisor Hsueh, Huey-Miin en_US dc.contributor.author (作者) 吳冠輝 zh_TW dc.contributor.author (作者) Wu, Guan-Hui en_US dc.creator (作者) 吳冠輝 zh_TW dc.creator (作者) Wu, Guan-Hui en_US dc.date (日期) 2019 en_US dc.date.accessioned 1-七月-2019 10:43:44 (UTC+8) - dc.date.available 1-七月-2019 10:43:44 (UTC+8) - dc.date.issued (上傳時間) 1-七月-2019 10:43:44 (UTC+8) - dc.identifier (其他 識別碼) G0106354027 en_US dc.identifier.uri (URI) http://nccur.lib.nccu.edu.tw/handle/140.119/124122 - dc.description (描述) 碩士 zh_TW dc.description (描述) 國立政治大學 zh_TW dc.description (描述) 統計學系 zh_TW dc.description (描述) 106354027 zh_TW dc.description.abstract (摘要) 由於中文文本中的詞彙之間沒有任何標記或空格，所以斷詞被認為是中文文本探勘前必要且重要的預處理步驟。而目前中文斷詞方法多屬監督式方法，當沒有適當的詞典時難以發揮，例如針對新世代的文章或特定專業領域的文本。Deng等人在2016年提出非監督式斷詞方法TopWORDS，利用文字詞典模型(Word D ictionary Model, WDM)建構文本之概似函數，並且將斷詞資訊當作遺失變數，以EM演算法估計出各詞彙的使用機率，更利用動態規劃法(dynamic programm ing)計算，除了運算上相當具有效率，TopWORDS應用在許多文本上有良好的結果。然而，TopWORDS假設文本中每個位置的詞彙獨立且分配相同，這樣的假設恐怕忽略了詞彙在文意上的相連。此研究假設每個詞彙出現的概率與前一個詞彙有關，因此文本的概似函數可表示為兩詞彙的序列關係的函數，故將此研究提出的方法稱為「SeqWORDS」。在運用三種不同斷詞法於紅樓夢文本上後，我們觀察到 SeqWORDS雖然在探索新詞彙的能力較弱，然而當接續使用文本探勘工具如詞向量分析後發現，SeqWORDS 能提供最佳的解釋性。 zh_TW dc.description.abstract (摘要) Unlike alphabet-based language, there exists no space between words inChinese corpus. The first step in Chinese text mining is to segment words in a sentence. Many existing segmentation methods are supervised in terms of requiring an adequate dictionary. However, Chinese language has developed so long and growing so fast. A suitable dictionary may not be available or easily accessed. In 2016, Deng et al. proposed an unsupervised method called “TopWORDS”, which needs no dictionary in hand. The authors derived the likelihood function of the corpus via word dictionary model (WDM). Further, they regard unknown segmentation information as missing data and utilize EM algorithm to estimate occurrence probability of words. To enhance computational efficiency, the estimates are computed by dynamic programming. In the article, the TopWORDS is found to perform well in several corpus. However, the iid assumption of TopWORDS ignores words dependency, which frequently occurs in consecutive words. Therefore, in this research we assume that a word’s occurrence depends on previous one and modify the TopWORDS method. By considering the sequential association of consecutive words, the proposed method is named “SeqWORDS”. The new method and two other existing methods are evaluated by their performance on the famous classical novel Story-of-Stone. We find that SeqWORDS is less capable to find new, rare words and is much time consuming. However, when we further implement some advance text mining analysis on the segmented corpus, the segmented corpus by SeqWORDS produces the most reasonable, interpretable results. en_US dc.description.tableofcontents 第一章 介紹 1第二章 方法 3第三章 實作 11第四章 結論 28參考文獻 30附錄一 33附錄二 35附錄三 38 zh_TW dc.format.extent 2540097 bytes - dc.format.mimetype application/pdf - dc.source.uri (資料來源) http://thesis.lib.nccu.edu.tw/record/#G0106354027 en_US dc.subject (關鍵詞) 中文斷詞 zh_TW dc.subject (關鍵詞) 文本探勘 zh_TW dc.subject (關鍵詞) 動態規劃法 zh_TW dc.subject (關鍵詞) 文字詞典模型 zh_TW dc.subject (關鍵詞) EM演算法 zh_TW dc.subject (關鍵詞) 詞彙序列關係 zh_TW dc.subject (關鍵詞) Chinese texts mining en_US dc.subject (關鍵詞) Dynamic programming en_US dc.subject (關鍵詞) EM algorithm en_US dc.subject (關鍵詞) Word dictionary model en_US dc.subject (關鍵詞) Words dependency en_US dc.subject (關鍵詞) Word segmentation en_US dc.title (題名) 基於兩詞彙的序列關係建造非監督式 SeqWORDS 斷詞方法 zh_TW dc.title (題名) SeqWORDS: an unsupervised Chinese segmentation method using relationship of two consecutive words. en_US dc.type (資料類型) thesis en_US dc.relation.reference (參考文獻) [1]The Stanford Natural Language. Processing Group, Chinese Natural Language Processing and Speech Processing. Retrieved May 24, 2019, from https://nlp.stanford.edu/projects/chinese-nlp.shtml#cws[2]J. Lafferty, A. McCallum, F. C.N. Pereira, (2001), Conditional random fields: Probabilistc models for segmenting and labeling sequence data. Proceedings of the 18th International Conference on Machine Learning 2001(ICML 2001), pp 282–289.[3]fxsjy, Jieba, Retrieved May 27, 2019, from https://github.com/fxsjy/ji eba[4]L. R. Rabiner, B. H. Juang, (1986), An introduction to hidden Markov models, IEEE ASSP MAGAZINE, vol 3, no 1, pp. 4-16.[5]A. Chen, (2003), Chinese word segmentation using minimal linguistic knowledge. Proceeding SIGHAN `03 Proceedings of the second SIGHAN workshop on Chinese language processing, Vol 17, pp 148–151.[6]K. J. Chen, S. H. Liu, (1992), Word identification for Mandarin Chinese sentences. Proceeding COLING `92 Proceedings of the 14th conference on Computational linguistics, Vol 1, pp 101–107.[7]K. Deng, P. K. Bol, K. J. Li, and J. S. Liu, (2016). On the unsupervised analysis of domain-specific Chinese texts. Proceedings of the National Academy of Sciences of the United States of America, vol 113, pp 6154–6159.[8]X. Ge, W. Pratt, P. Smyth, (1999), Discovering Chinese words from unsegmented text. Proceeding SIGIR `99 Proceedings of the 22nd annual international ACM SIGIR conference on Research and development in information retrieval, pp 271–272.[9]A. P. Dempster, N. M. Laird, D. B. Rubin, (1977), Maximum likelihood from incomplete data via the EM algorithm. Journal of the Royal Statistical Society. Series B, vol 39, no 1, pp 1-38.[10]R. Bellman, (1954), The theory of dynamic programming, Bulletin of the American Mathematical Society, vol 60, no 6, pp 503-515.[11]X. Cao, Story-of-Stone.[12] 胡適，(1988)，胡適紅樓夢研究論述全編，上海古籍出版社。[13]T. Mikolov, K. Chen, G. Corrado, J. Dean, (2013). Efficient Estimation of Word Representations in Vector Space, arXiv:1301.3781v3.[14]T. Mikolov, I. Sutskever, K. Chen, G. Corrado, J. Dean, (2013). DistributedRepresentations of Words and Phrases and their Compositionality, NIPS 2013,3111-3119.[15]K. Pearson, (1901), On Lines and Planes of Closest Fit to Systems of Points in Space. Philosophical Magazine, vol 2, pp 559-572. zh_TW dc.identifier.doi (DOI) 10.6814/NCCU201900115 en_US
