| { |
| "paper_id": "O16-1025", |
| "header": { |
| "generated_with": "S2ORC 1.0.0", |
| "date_generated": "2023-01-19T08:05:00.866448Z" |
| }, |
| "title": "\u547d\u540d\u5be6\u9ad4\u8b58\u5225\u904b\u7528\u65bc\u7522\u54c1\u540c\u7fa9\u8a5e\u64f4\u589e Using Named Entity Recognition Increases the Synonym of Products", |
| "authors": [ |
| { |
| "first": "\u6d2a\u667a\u529b", |
| "middle": [], |
| "last": "Chihli", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Yuan Christian University", |
| "location": {} |
| }, |
| "email": "chihli@cycu.edu.tw" |
| }, |
| { |
| "first": "Jheng-Hua", |
| "middle": [], |
| "last": "\u9ec3\u653f\u83ef", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Yuan Christian University", |
| "location": {} |
| }, |
| "email": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Huang", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Yuan Christian University", |
| "location": {} |
| }, |
| "email": "" |
| }, |
| { |
| "first": "Rui-Jia", |
| "middle": [], |
| "last": "\u937e\u745e\u5609", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Yuan Christian University", |
| "location": {} |
| }, |
| "email": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Zhong", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Yuan Christian University", |
| "location": {} |
| }, |
| "email": "" |
| }, |
| { |
| "first": "Liang-Pu", |
| "middle": [], |
| "last": "\u9673\u826f\u5703", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Yuan Christian University", |
| "location": {} |
| }, |
| "email": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Chen", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "Ping-Che", |
| "middle": [], |
| "last": "\u694a\u79c9\u54f2", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Yang", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| } |
| ], |
| "year": "", |
| "venue": null, |
| "identifiers": {}, |
| "abstract": "This research proposes the probability mapping approach for a product name to attempt to resolve the synonym identification problem, which is usually ignored in the field of Named Entity Recognition. Using the same name to describe a product or service may effectively improve the results of opinion mining or sentiment analysis. However, as WOM is a user generated content (UGC), different names may be used by the same or different users. Besides, there is no unified naming rule when writing the WOM. Even though the authors are the same or different, they may use different names to describe the same products. In this case, searching or organizing the WOM article without the consideration of the naming issue may lead to the problem of information loss. Thus, we propose the probability mapping approach via the co-occurrence naming dataset and the Word2vect language model in order to reduce the naming issue. According to our initially experimental results, the probability mapping approach for a product name present its potential in the naming issue.", |
| "pdf_parse": { |
| "paper_id": "O16-1025", |
| "_pdf_hash": "", |
| "abstract": [ |
| { |
| "text": "This research proposes the probability mapping approach for a product name to attempt to resolve the synonym identification problem, which is usually ignored in the field of Named Entity Recognition. Using the same name to describe a product or service may effectively improve the results of opinion mining or sentiment analysis. However, as WOM is a user generated content (UGC), different names may be used by the same or different users. Besides, there is no unified naming rule when writing the WOM. Even though the authors are the same or different, they may use different names to describe the same products. In this case, searching or organizing the WOM article without the consideration of the naming issue may lead to the problem of information loss. Thus, we propose the probability mapping approach via the co-occurrence naming dataset and the Word2vect language model in order to reduce the naming issue. According to our initially experimental results, the probability mapping approach for a product name present its potential in the naming issue.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Abstract", |
| "sec_num": null |
| } |
| ], |
| "body_text": [], |
| "back_matter": [], |
| "bib_entries": { |
| "BIBREF0": { |
| "ref_id": "b0", |
| "title": "Efficient Estimation of Word Representations in Vector Space", |
| "authors": [ |
| { |
| "first": "T", |
| "middle": [], |
| "last": "Mikolov", |
| "suffix": "" |
| }, |
| { |
| "first": "K", |
| "middle": [], |
| "last": "Chen", |
| "suffix": "" |
| }, |
| { |
| "first": "G", |
| "middle": [], |
| "last": "Corrado", |
| "suffix": "" |
| }, |
| { |
| "first": "J", |
| "middle": [], |
| "last": "Dean", |
| "suffix": "" |
| } |
| ], |
| "year": 2013, |
| "venue": "ICLR Workshop", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "T. Mikolov, K. Chen, G. Corrado, and J. Dean, \"Efficient Estimation of Word Representations in Vector Space,\" ICLR Workshop, Jan. 2013.", |
| "links": null |
| }, |
| "BIBREF1": { |
| "ref_id": "b1", |
| "title": "SentiMI: Introducing point-wise mutual information with SentiWordNet to improve sentiment polarity detection", |
| "authors": [ |
| { |
| "first": "F", |
| "middle": [ |
| "H" |
| ], |
| "last": "Khan", |
| "suffix": "" |
| }, |
| { |
| "first": "U", |
| "middle": [], |
| "last": "Qamar", |
| "suffix": "" |
| }, |
| { |
| "first": "S", |
| "middle": [], |
| "last": "Bashir", |
| "suffix": "" |
| } |
| ], |
| "year": 2016, |
| "venue": "Appl. Soft Comput", |
| "volume": "39", |
| "issue": "", |
| "pages": "140--153", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "F. H. Khan, U. Qamar, and S. Bashir, \"SentiMI: Introducing point-wise mutual information with SentiWordNet to improve sentiment polarity detection,\" Appl. Soft Comput., vol. 39, pp. 140-153, Feb. 2016.", |
| "links": null |
| }, |
| "BIBREF2": { |
| "ref_id": "b2", |
| "title": "Word of mouth and interpersonal communication: A review and directions for future research", |
| "authors": [ |
| { |
| "first": "J", |
| "middle": [], |
| "last": "Berger", |
| "suffix": "" |
| } |
| ], |
| "year": 2014, |
| "venue": "J. Consum. Psychol", |
| "volume": "24", |
| "issue": "4", |
| "pages": "586--607", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "J. Berger, \"Word of mouth and interpersonal communication: A review and directions for future research,\" J. Consum. Psychol., vol. 24, no. 4, pp. 586-607, Oct. 2014.", |
| "links": null |
| }, |
| "BIBREF3": { |
| "ref_id": "b3", |
| "title": "Opinion Observer: Analyzing and Comparing Opinions on the Web", |
| "authors": [ |
| { |
| "first": "B", |
| "middle": [], |
| "last": "Liu", |
| "suffix": "" |
| }, |
| { |
| "first": "M", |
| "middle": [], |
| "last": "Hu", |
| "suffix": "" |
| }, |
| { |
| "first": "J", |
| "middle": [], |
| "last": "Cheng", |
| "suffix": "" |
| } |
| ], |
| "year": 2005, |
| "venue": "Proceedings of the 14th International Conference on World Wide Web", |
| "volume": "", |
| "issue": "", |
| "pages": "342--351", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "B. Liu, M. Hu, and J. Cheng, \"Opinion Observer: Analyzing and Comparing Opinions on the Web,\" in Proceedings of the 14th International Conference on World Wide Web, New York, NY, USA, 2005, pp. 342-351.", |
| "links": null |
| }, |
| "BIBREF4": { |
| "ref_id": "b4", |
| "title": "Chinese Word Segmentation as Character Tagging", |
| "authors": [ |
| { |
| "first": "N", |
| "middle": [], |
| "last": "Xue", |
| "suffix": "" |
| } |
| ], |
| "year": 2003, |
| "venue": "Comput. Linguist. Chin. Lang. Process", |
| "volume": "8", |
| "issue": "1", |
| "pages": "29--48", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "N. Xue, \"Chinese Word Segmentation as Character Tagging,\" Comput. Linguist. Chin. Lang. Process., vol. 8, no. 1, pp. 29-48, Feb. 2003.", |
| "links": null |
| }, |
| "BIBREF5": { |
| "ref_id": "b5", |
| "title": "Latent semantics in Named Entity Recognition", |
| "authors": [ |
| { |
| "first": "M", |
| "middle": [], |
| "last": "Konkol", |
| "suffix": "" |
| }, |
| { |
| "first": "T", |
| "middle": [], |
| "last": "Brychc\u00edn", |
| "suffix": "" |
| }, |
| { |
| "first": "M", |
| "middle": [], |
| "last": "Konop\u00edk", |
| "suffix": "" |
| } |
| ], |
| "year": 2015, |
| "venue": "Expert Syst. Appl", |
| "volume": "42", |
| "issue": "7", |
| "pages": "3470--3479", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "M. Konkol, T. Brychc\u00edn, and M. Konop\u00edk, \"Latent semantics in Named Entity Recognition,\" Expert Syst. Appl., vol. 42, no. 7, pp. 3470-3479, May 2015.", |
| "links": null |
| }, |
| "BIBREF7": { |
| "ref_id": "b7", |
| "title": "Robust multilingual Named Entity Recognition with shallow semi-supervised features", |
| "authors": [ |
| { |
| "first": "R", |
| "middle": [], |
| "last": "Agerri", |
| "suffix": "" |
| }, |
| { |
| "first": "G", |
| "middle": [], |
| "last": "Rigau", |
| "suffix": "" |
| } |
| ], |
| "year": 2016, |
| "venue": "Artif. Intell", |
| "volume": "238", |
| "issue": "", |
| "pages": "63--82", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "R. Agerri and G. Rigau, \"Robust multilingual Named Entity Recognition with shallow semi-supervised features,\" Artif. Intell., vol. 238, pp. 63-82, Sep. 2016.", |
| "links": null |
| }, |
| "BIBREF8": { |
| "ref_id": "b8", |
| "title": "Privileged contextual information for context-aware recommender systems", |
| "authors": [ |
| { |
| "first": "C", |
| "middle": [ |
| "V" |
| ], |
| "last": "Sundermann", |
| "suffix": "" |
| }, |
| { |
| "first": "M", |
| "middle": [ |
| "A" |
| ], |
| "last": "Domingues", |
| "suffix": "" |
| }, |
| { |
| "first": "M", |
| "middle": [], |
| "last": "Da", |
| "suffix": "" |
| }, |
| { |
| "first": "S", |
| "middle": [], |
| "last": "Conrado", |
| "suffix": "" |
| }, |
| { |
| "first": "S", |
| "middle": [ |
| "O" |
| ], |
| "last": "Rezende", |
| "suffix": "" |
| } |
| ], |
| "year": 2016, |
| "venue": "Expert Syst. Appl", |
| "volume": "57", |
| "issue": "", |
| "pages": "139--158", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "C. V. Sundermann, M. A. Domingues, M. da S. Conrado, and S. O. Rezende, \"Privileged contextual information for context-aware recommender systems,\" Expert Syst. Appl., vol. 57, pp. 139-158, Sep. 2016.", |
| "links": null |
| }, |
| "BIBREF9": { |
| "ref_id": "b9", |
| "title": "Named Entity Recognition: Fallacies, challenges and opportunities", |
| "authors": [ |
| { |
| "first": "M", |
| "middle": [], |
| "last": "Marrero", |
| "suffix": "" |
| }, |
| { |
| "first": "J", |
| "middle": [], |
| "last": "Urbano", |
| "suffix": "" |
| }, |
| { |
| "first": "S", |
| "middle": [], |
| "last": "S\u00e1nchez-Cuadrado", |
| "suffix": "" |
| }, |
| { |
| "first": "J", |
| "middle": [], |
| "last": "Morato", |
| "suffix": "" |
| }, |
| { |
| "first": "J", |
| "middle": [ |
| "M" |
| ], |
| "last": "G\u00f3mez-Berb\u00eds", |
| "suffix": "" |
| } |
| ], |
| "year": 2013, |
| "venue": "Comput. Stand. Interfaces", |
| "volume": "35", |
| "issue": "5", |
| "pages": "482--489", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "M. Marrero, J. Urbano, S. S\u00e1nchez-Cuadrado, J. Morato, and J. M. G\u00f3mez-Berb\u00eds, \"Named Entity Recognition: Fallacies, challenges and opportunities,\" Comput. Stand. Interfaces, vol. 35, no. 5, pp. 482-489, Sep. 2013.", |
| "links": null |
| }, |
| "BIBREF10": { |
| "ref_id": "b10", |
| "title": "A hybrid named entity recognizer for Turkish", |
| "authors": [ |
| { |
| "first": "D", |
| "middle": [], |
| "last": "K\u00fc\u00e7\u00fck", |
| "suffix": "" |
| }, |
| { |
| "first": "A", |
| "middle": [], |
| "last": "Yaz\u0131c\u0131", |
| "suffix": "" |
| } |
| ], |
| "year": 2012, |
| "venue": "Expert Syst. Appl", |
| "volume": "39", |
| "issue": "3", |
| "pages": "2733--2742", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "D. K\u00fc\u00e7\u00fck and A. Yaz\u0131c\u0131, \"A hybrid named entity recognizer for Turkish,\" Expert Syst. Appl., vol. 39, no. 3, pp. 2733-2742, Feb. 2012.", |
| "links": null |
| }, |
| "BIBREF11": { |
| "ref_id": "b11", |
| "title": "Don't count, predict! A systematic comparison of context-counting vs. context-predicting semantic vectors", |
| "authors": [ |
| { |
| "first": "M", |
| "middle": [], |
| "last": "Baroni", |
| "suffix": "" |
| }, |
| { |
| "first": "G", |
| "middle": [], |
| "last": "Dinu", |
| "suffix": "" |
| } |
| ], |
| "year": 2014, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "M. Baroni and G. Dinu, \"Don't count, predict! A systematic comparison of context-counting vs. context-predicting semantic vectors,\" in In ACL, 2014.", |
| "links": null |
| }, |
| "BIBREF12": { |
| "ref_id": "b12", |
| "title": "Deep learning for visual understanding: A review", |
| "authors": [ |
| { |
| "first": "Y", |
| "middle": [], |
| "last": "Guo", |
| "suffix": "" |
| }, |
| { |
| "first": "Y", |
| "middle": [], |
| "last": "Liu", |
| "suffix": "" |
| }, |
| { |
| "first": "A", |
| "middle": [], |
| "last": "Oerlemans", |
| "suffix": "" |
| }, |
| { |
| "first": "S", |
| "middle": [], |
| "last": "Lao", |
| "suffix": "" |
| }, |
| { |
| "first": "S", |
| "middle": [], |
| "last": "Wu", |
| "suffix": "" |
| }, |
| { |
| "first": "M", |
| "middle": [ |
| "S" |
| ], |
| "last": "Lew", |
| "suffix": "" |
| } |
| ], |
| "year": 2016, |
| "venue": "Neurocomputing", |
| "volume": "187", |
| "issue": "", |
| "pages": "27--48", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Y. Guo, Y. Liu, A. Oerlemans, S. Lao, S. Wu, and M. S. Lew, \"Deep learning for visual understanding: A review,\" Neurocomputing, vol. 187, pp. 27-48, Apr. 2016.", |
| "links": null |
| }, |
| "BIBREF13": { |
| "ref_id": "b13", |
| "title": "Query-oriented unsupervised multi-document summarization via deep learning model", |
| "authors": [ |
| { |
| "first": "S", |
| "middle": [], |
| "last": "Zhong", |
| "suffix": "" |
| }, |
| { |
| "first": "Y", |
| "middle": [], |
| "last": "Liu", |
| "suffix": "" |
| }, |
| { |
| "first": "B", |
| "middle": [], |
| "last": "Li", |
| "suffix": "" |
| }, |
| { |
| "first": "J", |
| "middle": [], |
| "last": "Long", |
| "suffix": "" |
| } |
| ], |
| "year": 2015, |
| "venue": "Expert Syst. Appl", |
| "volume": "42", |
| "issue": "21", |
| "pages": "8146--8155", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "S. Zhong, Y. Liu, B. Li, and J. Long, \"Query-oriented unsupervised multi-document summarization via deep learning model,\" Expert Syst. Appl., vol. 42, no. 21, pp. 8146-8155, Nov. 2015.", |
| "links": null |
| }, |
| "BIBREF14": { |
| "ref_id": "b14", |
| "title": "Distributed Representations of Words and Phrases and their Compositionality", |
| "authors": [ |
| { |
| "first": "T", |
| "middle": [], |
| "last": "Mikolov", |
| "suffix": "" |
| }, |
| { |
| "first": "I", |
| "middle": [], |
| "last": "Sutskever", |
| "suffix": "" |
| }, |
| { |
| "first": "K", |
| "middle": [], |
| "last": "Chen", |
| "suffix": "" |
| }, |
| { |
| "first": "G", |
| "middle": [], |
| "last": "Corrado", |
| "suffix": "" |
| }, |
| { |
| "first": "J", |
| "middle": [], |
| "last": "Dean", |
| "suffix": "" |
| } |
| ], |
| "year": 2013, |
| "venue": "NIPS", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "T. Mikolov, I. Sutskever, K. Chen, G. Corrado, and J. Dean, \"Distributed Representations of Words and Phrases and their Compositionality,\" in NIPS, 2013.", |
| "links": null |
| }, |
| "BIBREF15": { |
| "ref_id": "b15", |
| "title": "Chinese comments sentiment classification based on word2vec and SVMperf", |
| "authors": [ |
| { |
| "first": "D", |
| "middle": [], |
| "last": "Zhang", |
| "suffix": "" |
| }, |
| { |
| "first": "H", |
| "middle": [], |
| "last": "Xu", |
| "suffix": "" |
| }, |
| { |
| "first": "Z", |
| "middle": [], |
| "last": "Su", |
| "suffix": "" |
| }, |
| { |
| "first": "Y", |
| "middle": [], |
| "last": "Xu", |
| "suffix": "" |
| } |
| ], |
| "year": 2015, |
| "venue": "Expert Syst. Appl", |
| "volume": "42", |
| "issue": "4", |
| "pages": "1857--1863", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "D. Zhang, H. Xu, Z. Su, and Y. Xu, \"Chinese comments sentiment classification based on word2vec and SVMperf,\" Expert Syst. Appl., vol. 42, no. 4, pp. 1857-1863, Mar. 2015.", |
| "links": null |
| } |
| }, |
| "ref_entries": { |
| "TABREF0": { |
| "type_str": "table", |
| "html": null, |
| "text": "\u7d71\u8a08\u6cd5(count-based method)\uff0c\u5982\u6f5b\u5728\u8a9e\u610f\u5206\u6790(LSA; latent semantic analysis)\u548c\u9810\u6e2c\u6cd5 (predictive method) \uff0c \u5982 \u985e \u795e \u7d93 \u6a5f \u7387 \u8a9e \u8a00 \u6a21 \u578b (NPLM; neural probabilistic language model)\u3002\u8a5e\u983b\u7d71\u8a08\u6cd5\u5f9e\u5927\u91cf\u6587\u5b57\u8cc7\u6599\u4e2d\u7d71\u8a08\u76ee\u6a19\u5b57\u548c\u9130\u8fd1\u5b57\u7684\u5171\u73fe\u95dc\u4fc2\uff0c\u9810\u6e2c\u6cd5\u85c9\u7531\u9130 Word2vec \u70ba Google \u65bc 2013 \u91cb\u51fa\u4e4b\u958b\u6e90\u5de5\u5177\u4e4b\u4e00\uff0c\u6839\u64da Mikolov et al. (2013) \u767c\u8868 \u8868\u53e3\u7891\u6587\u7ae0\u4e2d i \u7522\u54c1\u7684\u7e3d\u5b57\u6578\uff0cp j \u8868\u7522\u54c1\u8a5e\u5eab\u4e2d j \u7522\u54c1\u7684\u7e3d\u5b57\u6578\uff0cw i \u2229p j \u8868\u76f8\u540c\u7684", |
| "num": null, |
| "content": "<table><tr><td>\u7db2\u969b\u7db2\u8def\u7684\u767c\u5c55\uff0c\u9020\u5c31\u4e86\u6d88\u8cbb\u8005\u5c0d\u65bc\u8868\u9054\u610f\u898b\u7684\u65b9\u5f0f\u6709\u5de8\u5927\u7684\u6539\u8b8a\u3002\u6d88\u8cbb\u8005\u5f9e\u88ab\u52d5\u7684\u63a5 \u6536\u8a0a\u606f\uff0c\u8f49\u8b8a\u6210\u4e3b\u52d5\u5c0d\u65bc\u7522\u54c1\u7684\u4f7f\u7528\u7d93\u9a57\uff0c\u5728\u5404\u7a2e\u7db2\u969b\u7db2\u8def\u7684\u5e73\u53f0\u4e0a\uff0c\u767c\u8868\u4e26\u5206\u4eab\u76f8\u95dc \u7684\u8a55\u8ad6[2]\u3002\u7db2\u969b\u7db2\u8def\u4e0a\u6240\u6563\u4f48\u7684\u96fb\u5b50\u53e3\u7891\u5177\u6709\u5de8\u5927\u884c\u92b7\u50f9\u503c\uff0c\u6d88\u8cbb\u8005\u53ef\u4ee5\u900f\u904e\u7db2\u8def\u53e3 \u547d\u540d\u5be6\u9ad4B\u662f\u5426\u76f8\u540c\u3002\u56e0\u6b64\uff0c\u6587\u737b\u4e0a\uff0c\u4ecd\u7136\u7f3a\u4e4f\u5224\u65b7\u547d\u540d\u5be6\u9ad4\u9593\u7684\u540c\u7fa9\u8a5e\u95dc\u4fc2\u7684\u76f8\u95dc\u7814 \u7a76\uff0c\u672c\u7814\u7a76\u5247\u5c08\u6ce8\u65bc\u547d\u540d\u5be6\u9ad4\u7684\u540c\u7fa9\u8a5e\u7814\u7a76\uff0c\u9664\u4e86\u6539\u5584\u5728\u547d\u540d\u5be6\u9ad4\u4e2d\u6240\u7522\u751f\u7684\u8cc7\u8a0a\u907a\u6f0f \u7684\u554f\u984c\uff0c\u4e5f\u5617\u8a66\u586b\u88dc\u5728\u6b64\u7814\u7a76\u9818\u57df\u6587\u737b\u4e0a\u7684\u4e0d\u8db3\u3002 \u4e09\u3001 \u7522\u54c1\u540d\u7a31\u6a5f\u7387\u6bd4\u5c0d\u6cd5 \u672c\u7814\u7a76\u4e2d\u63d0\u51fa\u7684\u7522\u54c1\u540d\u7a31\u6a5f\u7387\u6bd4\u5c0d\u6cd5\u64f4\u589e\u540c\u7fa9\u8a5e\u8a5e\u5eab\u6a21\u578b\uff0c\u5206\u5225\u70ba\u624b\u52d5\u64f4\u589e\u548c\u81ea\u52d5 \u5316\u8f38\u5165\u3002\u624b\u52d5\u64f4\u589e\u5247\u70ba\u4ee5\u4eba\u5de5\u65b9\u5f0f\u5efa\u7acb\u5df2\u77e5\u5546\u54c1\u540c\u7fa9\u8a5e\u8a5e\u5eab\uff0c\u81ea\u52d5\u5316\u64f4\u589e\u5982\u4e0b\u5716\u4e00\uff0c\u900f \u5b9a\u9580\u6abb\u503c(\u5982 60%)\uff0c\u5247\u6b64\u53e3\u7891\u6587\u7ae0\u548c\u76ee\u6a19\u7522\u54c1\u8a5e\u5f59\u5177\u5099\u9ad8\u5ea6\u95dc\u4fc2\u3002\u5982\u67d0\u53e3\u7891\u6587\u7ae0\u65b7\u8a5e\u5f8c \u56db\u3001 \u5be6\u9a57 \u672c\u7814\u7a76\u9664\u4e86\u4ee5\u300c\u7fa9\u7f8e\u8c46\u5976\u300d\u505a\u70ba\u5be6\u9a57\u7684\u7522\u54c1\u8a5e\u5f59\uff0c\u4e5f\u91dd\u5c0d\u738b\u5b50\u9eb5\u3001\u9ea5\u9999\u7d05\u8336\u3001\u53ef\u53e3 \u516d\u3001 \u81f4\u8b1d Chinese text segmentation, https://github.com/fxsjy/jieba) \u65b7\u8a5e\u3002Jieba \u65b7\u8a5e\u7cfb\u7d71\u6709\u4e09\u7a2e\u65b7\u8a5e \u65b9\u5f0f\uff0c\u5206\u5225\u70ba\u7cbe\u78ba\u6a21\u5f0f\u3001\u5168\u6a21\u5f0f\u3001\u641c\u5c0b\u5f15\u64ce\u6a21\u5f0f\u3002\u7531\u65bc\u672c\u7814\u7a76\u9700\u8981\u9032\u884c\u8a5e\u5f59\u7684\u8a5e\u6027\u7be9\u9078\uff0c \u7684\u53e3\u7891\u8a5e\u5f59(\u5168\u5bb6 \u6709 \u7121\u7cd6 \u7fa9\u7f8e\u8c46\u6f3f \u597d\u559d)\uff0c\u5728\u5171\u73fe\u8a5e\u5f59\u96c6\u4e2d\u767c\u73fe\uff0c \u300c\u7121\u7cd6\u300d \u3001 \u300c\u597d\u559d\u300d \u3001 (\u4e00)\u3001 \u5be6\u9a57\u8a2d\u8a08 \u53ef\u6a02\u4e09\u7a2e\u7522\u54c1\u9032\u884c\u540c\u7fa9\u8a5e\u64f4\u5145\u5be6\u9a57\uff0c\u5176\u53ef\u80fd\u7684\u540c\u7fa9\u8a5e\u5982\u8868\u56db\u6240\u793a\u3002 \u672c\u7814\u7a76\u4f9d\u7d93\u6fdf\u90e8\u88dc\u52a9\u8ca1\u5718\u6cd5\u4eba\u8cc7\u8a0a\u5de5\u696d\u7b56\u9032\u6703 \u300c105 \u5e74\u5ea6\u865b\u5be6\u6574\u5408\u667a\u6167\u5546\u52d9\u95dc\u9375\u6280 \u300c\u5168\u5bb6\u300d\u5747\u548c\u300c\u7fa9\u7f8e\u8c46\u5976\u300d\u6709\u9ad8\u5ea6\u5171\u73fe\u95dc\u4fc2\uff0c\u5247\u6b64\u53e3\u7891\u6587\u7ae0\u6240\u63cf\u7e6a\u7684\u7522\u54c1\u53ef\u63a8\u5b9a\u70ba\u300c\u7fa9 \u56e0\u6b64\u63a1\u7528 Jieba \u7cbe\u78ba\u65b7\u8a5e\u6a21\u5f0f\u3002\u7d93\u591a\u6b21\u5be6\u9a57\uff0c\u672c\u7814\u7a76\u6700\u5f8c\u4fdd\u7559\u7684\u8a5e\u6027\u70ba\u540d\u8a5e\u3001\u52d5\u8a5e\u548c\u5f62 \u7f8e\u8c46\u5976\u300d \uff0c\u6b64\u6642\u6703\u5c07\u53e3\u7891\u4e2d\u7684\u8a5e\u5f59(\u5168\u5bb6\u3001\u6709\u3001\u7121\u7cd6\u3001\u7fa9\u7f8e\u8c46\u6f3f\u3001\u597d\u559d)\u8996\u70ba\u53ef\u80fd\u7684\u540c\u7fa9 \u672c\u7814\u7a76\u7684\u5be6\u9a57\u5c1a\u8655\u65bc\u6e2c\u8a66\u968e\u6bb5\uff0c\u76ee\u524d\u8cc7\u6599\u6578\u91cf\u70ba 2141 \u7b46\u53e3\u7891\u8cc7\u6599\uff0c\u8cc7\u6599\u96c6\u6240\u5305\u542b \u8868\u56db\u3001\u76f8\u4f3c\u8a5e\u5f59\u8868 \u8853\u8207\u5e73\u53f0\u7814\u767c\u8a08\u756b(3/4)\u300d\u8fa6\u7406\u3002</td></tr><tr><td>\u7891\uff0c\u505a\u51fa\u8cfc\u8cb7\u8207\u5426\u7684\u6c7a\u5b9a[3]\uff1b\u5ee0\u5546\u53ef\u4ee5\u7531\u6d41\u50b3\u7db2\u8def\u7684\u53e3\u7891\uff0c\u5206\u6790\u6d88\u8cbb\u8005\u5c0d\u65bc\u6240\u5c6c\u7522\u54c1 \u7684\u610f\u898b\uff0c\u4f5c\u70ba\u7522\u54c1\u6539\u9032\u7684\u91cd\u8981\u53c3\u8003[4]\u3002\u53e3\u7891\u610f\u898b\u4ee5\u975e\u7d50\u69cb\u5316\u7684\u6587\u5b57\u8cc7\u6599\u5448\u73fe\uff0c\u8f49\u5316\u9019 \u4e9b\u975e\u7d50\u69cb\u5316\u7684\u5927\u91cf\u8cc7\u6599\u70ba\u5ee0\u5546\u548c\u7db2\u6c11\u53ef\u4ee5\u76f4\u63a5\u5229\u7528\u7684\u77e5\u8b58\uff0c\u4e00\u76f4\u70ba\u610f\u898b\u63a2\u52d8\u6216\u60c5\u611f\u5206\u6790 \u9818\u57df\u7684\u4e00\u9805\u91cd\u8981\u6311\u6230[2]\u3002\u7522\u54c1\u540d\u7a31\u4e0d\u4e00\u81f4\u7684\u554f\u984c\uff0c\u52a0\u6df1\u6b64\u6311\u6230\u7684\u56f0\u96e3\u5ea6\uff0c\u4e5f\u8b93\u610f\u898b\u63a2 \u52d8\u6210\u70ba\u76f8\u95dc\u5b78\u8853\u754c\u548c\u5be6\u52d9\u754c\u5171\u540c\u95dc\u6ce8\u7684\u4e00\u9805\u91cd\u8981\u8b70\u984c\u3002 \u5728\u53f0\u7063\u5e38\u4f7f\u7528\u7684\u53e3\u7891\u8cc7\u6599\u53ef\u5206\u70ba\u4e2d\u6587\u6216\u82f1\u6587\uff0c\u672c\u7814\u7a76\u9396\u5b9a\u4e2d\u6587\u53e3\u7891\u3002\u4e2d\u6587\u53e3\u7891\u548c\u82f1 \u6587\u53e3\u7891\u6700\u5927\u7684\u4e0d\u540c\u8655\u5728\u65bc\uff0c\u4e2d\u6587\u4e26\u7121\u81ea\u7136\u5b58\u5728\u7684\u8a5e\u9593\u65b7\u8a5e\u7b26\u865f\uff0c\u800c\u82f1\u6587\u7684\u7a7a\u767d\u5373\u70ba\u8a5e\u8207 \u8a5e\u4e4b\u9593\u7684\u5206\u9694\u4f9d\u64da\u3002\u55ae\u4e00\u4e2d\u6587\u65b9\u584a\u5b57(Chinese character)\u7684\u5b57\u7fa9\u904e\u65bc\u6a21\u7cca\uff0c\u7121\u6cd5\u8868\u9054\u5b8c\u6574 \u7684\u6982\u5ff5\uff0c\u5982\u8f38\u8d0f\u7684\u300c\u8f38\u300d\u548c\u904b\u8f38\u7684\u300c\u8f38\u300d\uff0c\u5b57\u578b\u76f8\u540c\u5b57\u7fa9\u537b\u622a\u7136\u4e0d\u540c\u3002\u6240\u4ee5\u5c0d\u65bc\u4e2d\u6587\u7684 \u6587\u5b57\u8655\u7406\uff0c\u9700\u8981\u518d\u984d\u5916\u7684\u8655\u7406\u4e2d\u6587\u7684\u65b7\u8a5e\u554f\u984c[5]\u3002\u53e6\u4e00\u65b9\u9762\uff0c\u53e3\u7891\u6587\u7ae0\u70ba\u4f7f\u7528\u8005\u6240\u7522 \u751f\u7684\u5167\u5bb9(UGC; user generated content)\uff0c\u64b0\u5beb\u98a8\u683c\u8207\u7528\u8a5e\u7528\u8a9e\u4e0d\u53d7\u898f\u7bc4\uff0c\u7121\u6cd5\u907f\u514d\u7e2e\u5beb \u4e8c\u3001 \u6587\u737b\u63a2\u8a0e (\u4e00)\u3001 \u547d\u540d\u5be6\u9ad4\u8b58\u5225 \u547d\u540d\u5be6\u9ad4\u8b58\u5225\u4e3b\u8981\u7528\u65bc\u8fa8\u8b58\u540d\u7a31\u7279\u5fb5\u7684\u8868\u9054\u65b9\u5f0f\uff0c\u9019\u4e9b\u7279\u5fb5\u53ef\u4ee5\u662f\u4eba\u540d\u3001\u5730\u540d[8]\u3002 \u9664\u4e86\u540d\u7a31\u55ae\u4f4d\u4e4b\u5916\uff0c\u6578\u5b57\u7684\u8868\u9054\u8fa8\u8b58\u4e5f\u662f\u7814\u7a76\u7684\u7bc4\u570d\uff0c\u5982\u6642\u9593\u3001\u65e5\u671f\u3001\u8ca8\u5e63\u2026\u7b49\u7b49[9]\u3002 \u547d\u540d\u5be6\u9ad4\u8b58\u5225\u662f\u5c07\u8cc7\u6599\u9032\u884c\u8cc7\u8a0a\u8403\u53d6\uff0c\u5176\u4e3b\u8981\u7684\u529f\u80fd\u5305\u62ec\u8b58\u5225\u548c\u5206\u985e\u67d0\u4e9b\u7a2e\u985e\u7684\u8cc7\u8a0a\u5143 \u7d20\u540d\u7a31\u3002\u56e0\u6b64\uff0c\u5176\u7d50\u679c\u53ef\u4ee5\u4f5c\u70ba\u8a9e\u7fa9\u6a19\u8a3b\u3001\u672c\u9ad4\u7684\u5efa\u69cb\u7b49\u61c9\u7528\u3002\u540c\u6642\u547d\u540d\u5be6\u9ad4\u8b58\u5225\u4e5f\u662f \u610f\u898b\u63a2\u52d8\u7684\u57fa\u790e\uff0c\u5728\u53e3\u7891\u610f\u898b\u4e2d\u900f\u904e\u547d\u540d\u5be6\u9ad4\u7684\u6280\u8853\u53ef\u4ee5\u6539\u5584\u610f\u898b\u7684\u5339\u914d\u7d50\u679c[10]\u3002 \u547d\u540d\u5be6\u9ad4\u8b58\u5225\u5728\u81ea\u7136\u8a9e\u8a00\u8655\u7406\u4e2d\u767c\u5c55\u8a31\u4e45\uff0c\u547d\u540d\u5be6\u9ad4\u8b58\u5225\u4e3b\u8981\u5206\u70ba\u4ee5\u898f\u5247\u70ba\u4e3b\u7684\u8fa8 \u8b58[11]\u3001\u6216\u662f\u4ee5\u6a5f\u5668\u5b78\u7fd2\u6cd5\u70ba\u4e3b\u7684\u8fa8\u8b58\u3002\u5e38\u898b\u4f7f\u7528\u65bc\u547d\u540d\u5be6\u9ad4\u8b58\u5225\u7684\u6a5f\u5668\u5b78\u7fd2\u6cd5\uff0c\u5982\u96b1 \u99ac\u53ef\u592b\u6a21\u578b(HMM; hidden Markov model)\u3001\u6c7a\u7b56\u6a39(DT; decision tree)\u3001\u6700\u5927\u71b5\u652f\u63f4\u5411\u91cf \u6a5f(MESVM; maximum entropy support vector machine)\u3002\u547d\u540d\u5be6\u9ad4\u8b58\u5225\u6240\u8fa8\u8b58\u7684\u6587\u5b57\u8cc7\u8a0a \u8fd1\u5b57\u9810\u6e2c\u76ee\u6a19\u5b57\u3002 \u6a5f\u5668\u5b78\u7fd2\u65b9\u6cd5\u4e3b\u8981\u904b\u7528\u96fb\u8166\u6a21\u64ec\u65b9\u5f0f\uff0c\u63a1\u7528\u6f14\u7b97\u6cd5\u5247\uff0c\u5176\u4e2d\u6df1\u5ea6\u5b78\u7fd2(DP; deep learning)\u5728 AlphaGo \u6a5f\u5668\u4eba\u7a0b\u5f0f\u9023\u7e8c\u6311\u6230\u97d3\u570b\u570d\u68cb\u68cb\u738b\u6210\u529f\u5f8c\u8072\u540d\u5927\u566a\u3002\u6df1\u5ea6\u5b78\u7fd2\u76ee\u524d \u5df2\u4f7f\u7528\u65bc\u8a08\u7b97\u6a5f\u8996\u89ba\u7684\u7814\u7a76\u9818\u57df\u4e2d\uff0c\u5982\u5f71\u50cf\u5206\u985e\u3001\u76ee\u6a19\u5075\u6e2c\u3001\u5f71\u50cf\u6aa2\u7d22\uff0c\u5176\u5b78\u7fd2\u6a21\u578b\u53ef \u4ee5\u5206\u70ba\u56db\u985e\uff0c\u5982\u9650\u5236\u6ce2\u723e\u8332\u66fc\u6a5f\u5668(restricted Boltzmann machine)\u3001\u7d30\u80de\u5f0f\u985e\u795e\u7d93\u7db2\u8def (cellular neural network)\u3001\u81ea\u7de8\u78bc\u5668(autoencoders)\u3001\u7a00\u758f\u6027\u5806\u758a\u81ea\u7de8\u78bc\u5668(stacked sparse \u5927\u91cf\u7684\u8cc7\u6599\u5f80\u5f80\u80fd\u8b93\u7c21\u55ae\u7684\u6f14\u7b97\u6a21\u578b\u7684\u6548\u679c\uff0c\u9ad8\u65bc\u53ea\u4f7f\u7528\u8f03\u5c11\u8cc7\u6599\u4f46\u537b\u8a2d\u8a08\u8907\u96dc\u7684\u6f14\u7b97 \u6a21\u578b[1]\u3002\u6df1\u5ea6\u5b78\u7fd2\u9664\u4e86\u53ef\u4ee5\u904b\u7528\u65bc\u8a08\u7b97\u6a5f\u8996\u89ba\u4e2d\uff0c\u4e5f\u53ef\u7528\u65bc\u6587\u5b57\u8cc7\u6599\u4e4b\u4e2d\uff0c\u5982\u7528\u65bc\u6587 \u5b57\u8cc7\u6599\u7684\u6458\u8981\u8403\u53d6[14]\u3002 \u58c7\u6536\u96c6\u7684\u53e3\u7891\u6587\u7ae0\u4e2d\uff0c\u7d93\u904e\u8cc7\u6599\u524d\u8655\u7406\u6b65\u9a5f\uff0c\u4f9d\u7167\u5b57\u8a5e\u51fa\u73fe\u6b21\u6578\uff0c\u5efa\u7acb\u5171\u73fe\u8a5e\u5f59\u96c6\uff0c\u5982 \u8868\u4e00\u3002\u7522\u54c1\u8a5e\u5f59\u7d44\u5408\u7684\u76f8\u4f3c\u5ea6\u6703\u900f\u904e\u516c\u5f0f\u4e00\uff0c\u8a08\u7b97\u9918\u5f26\u76f8\u4f3c\u5ea6\uff0c\u5728\u6b64\u7684\u8a5e\u5f59\u76f8\u4f3c\u5ea6\u7d44\u5408 \u50c5\u4ee5\u7522\u54c1\u8a5e\u5f59\u8207\u53e3\u7891\u8a5e\u5f59\u9032\u884c\u904b\u7b97\uff0c\u4e0d\u9032\u884c\u53e3\u7891\u8a5e\u5f59\u9593\u7684\u904b\u7b97\uff0c\u5982\u904b\u7b97\u300c\u7fa9\u7f8e\u8c46\u5976\u300d\u8207 \u300c\u967d\u5149\u300d \u3001 \u300c\u7fa9\u7f8e\u8c46\u5976\u300d\u8207\u300c\u7a40\u7269\u300d\u7b49\u7684\u9918\u5f26\u76f8\u4f3c\u5ea6\uff0c\u4e0d\u8a08\u7b97\u300c\u967d\u5149\u300d\u8207\u300c\u597d\u559d\u300d\u7b49\u53e3\u7891 \u8a5e\u5f59\u9593\u7684\u9918\u5f26\u76f8\u4f3c\u5ea6\u3002\u56e0\u6b64\uff0c\u6211\u5011\u53ef\u4ee5\u5f97\u5230\u6240\u6709\u53e3\u7891\u8a5e\u5f59\u548c\u7522\u54c1\u8a5e\u5f59\u7684\u5171\u73fe\u503c\u3002 \u7684\u8a13\u7df4\uff1aCBOW \u548c skip-gram\uff0c\u7531\u65bc skip-gram \u9069\u5408\u7528\u4f86\u8655\u7406\u5927\u91cf\u8cc7\u6599\uff0c\u672c\u7814\u7a76\u7684\u521d\u671f \u5be6\u9a57\u63a1\u7528\u7684\u65b9\u5f0f\u70ba skip-gram\uff0c\u5176\u985e\u795e\u7d93\u7db2\u8def\u793a\u610f\u5716\uff0c\u5982\u5716\u4e8c\uff0c\u80fd\u5920\u6839\u64da\u76ee\u6a19\u8a5e\u5f59\u548c\u5176 \u524d\u5f8c\u8a5e\u5f59\u7684\u95dc\u4fc2\u5efa\u7acb\u6a21\u578b\uff0c\u7576\u8f38\u5165\u76ee\u6a19\u8a5e\u5f59\u6642\uff0c\u7522\u751f\u548c\u76ee\u6a19\u8a5e\u5f59\u6700\u5177\u95dc\u4fc2\u7684\u524d\u5f8c\u8a5e\u5f59\u3002 \u6b64\u65b9\u6cd5\u548c\u672c\u7814\u7a76\u63d0\u51fa\uff0c\u5229\u7528\u5171\u73fe\u8a5e\u5f59\u96c6\u6bd4\u5c0d\u7684\u65b9\u5f0f\u4e0d\u540c\uff0c\u6703\u8003\u616e\u7522\u54c1\u8a5e\u5f59\u548c\u53e3\u7891\u8a5e\u5f59\u9593 (\u516d)\u3001 \u7522\u54c1\u540d\u7a31\u6bd4\u5c0d \u672c\u7814\u7a76\u6240\u63d0\u51fa\u7684\u7522\u54c1\u540d\u7a31\u6a5f\u7387\u6bd4\u5c0d\u6cd5\u70ba\uff0c\u8f38\u5165\u4e00\u7522\u54c1\u540d\u7a31\u5f8c\uff0c\u9032\u5165\u5230\u6bd4\u5c0d\u7684\u904e\u7a0b\uff0c \u904e\u7a0b\u4e2d\u6703\u5c07\u5df2\u65b7\u8a5e\u7684\u53e3\u7891\u6587\u7ae0\u8207\u7522\u54c1\u540d\u7a31\u9032\u884c\u6bd4\u5c0d\u3002\u900f\u904e\u6b64\u6bd4\u5c0d\u65b9\u6cd5\u6703\u627e\u5c0b\u76f8\u540c\u6216\u76f8\u4f3c \u7684\u540d\u7a31\uff0c\u5982\u679c\u627e\u5230\u76f8\u540c\u7684\u7522\u54c1\u540d\u7a31\uff0c\u5c31\u6703\u8f38\u51fa\u53e3\u7891\u6587\u7ae0\u4e26\u7d50\u675f\u6bd4\u5c0d\u6d41\u7a0b\u3002\u5982\u679c\u627e\u5230\u76f8\u4f3c \u53c3\u6578\u503c \u53c3\u6578\u6a21\u5f0f cbow =0 \u8f38\u51fa\u8a5e\u5411\u91cf\u7684\u7dad\u5ea6 \u8a13\u7df4\u6642\u5305\u542b\u524d\u5f8c\u6587\u7684\u9577\u5ea6 Hs= 1 \u8fed\u4ee3\u8a13\u7df4\u56de\u6578 \u53ef\u53e3\u53ef\u6a02 \u53ef\u6a02 0.5 iter =10 \u9ea5\u9999\u7d05\u8336 \u9ea5\u9999\u5976\u8336 0.5625 \u4f7f\u7528 Hierarchical Softmax \u6700\u4f73\u5316 \u9ea5\u9999\u7d05\u8336 \u9ea5\u9999\u7d05 0.75 Window= 5 \u738b\u5b50\u9eb5 \u5c0f\u738b\u5b50\u9eb5 0.75 Size= 400 \u7fa9\u7f8e\u8c46\u5976 \u7fa9\u7f8e\u8c46\u6f3f 0.5625 \u4f7f\u7528 Skip-gram \u6a21\u578b \u7522\u54c1\u540d\u7a31 \u540c\u7fa9\u8a5e \u76f8\u4f3c\u503c Word2vec \u6a21\u578b\u5c6c\u65bc\u6df1\u5ea6\u5b78\u7fd2\u7684\u4e00\u7a2e\u61c9\u7528\uff0c\u8a13\u7df4\u8a5e\u5f59\u6642\u53ef\u4ee5\u9078\u64c7\u5169\u7a2e\u6a21\u5f0f\u9032\u884c\u8a5e\u5f59 \u8868\u4e8c\u3001Word2vec \u53c3\u6578\u8868 \u8868\u4e94\u3001\u7522\u54c1\u540d\u7a31\u6bd4\u5c0d\u76f8\u4f3c\u8a5e\u5f59 autoencoder)\uff0c\u6df1\u5ea6\u5b78\u7fd2\u7684\u6a21\u578b\u70ba\u985e\u795e\u7d93\u7db2\u8def\u7684\u4e00\u7a2e\u884d\u751f\u6a21\u578b[13]\uff0c\u4e3b\u8981\u7684\u8ad6\u9ede\u5728\u65bc\u63a1\u7528 \u904e\u5171\u73fe\u8a5e\u5f59\u96c6\u3001Word2vec\u3001\u7522\u54c1\u540d\u7a31\u6bd4\u5c0d\uff0c\u5206\u5225\u6558\u8ff0\u5982\u4e0b\u3002 \u5bb9\u8a5e\uff0c\u56e0\u70ba\u5728\u4e2d\u6587\u53e3\u7891\u4e2d\u6b64\u4e09\u985e\u8a5e\u5f59\u548c\u547d\u540d\u5be6\u9ad4\u95dc\u4fc2\u6700\u70ba\u5bc6\u5207\u3002Jieba \u65b7\u8a5e\u7d50\u679c\u4ecd\u6709\u904e \u65bc\u96f6\u788e\u7684\u554f\u984c\uff0c\u5c0d\u65bc\u76ee\u6a19\u8a5e\u5f59\uff0c\u5247\u63a1\u53d6\u5b8c\u6574\u4fdd\u7559\u7b56\u7565\uff0c\u4ea6\u5373\u4e0d\u65b7\u8a5e\uff0c\u5982\u76ee\u6a19\u8a5e\u5f59\u70ba\u300c\u7fa9 \u7f8e\u8c46\u5976\u300d \uff0c\u5247\u78ba\u4fdd\u4e0d\u6703\u5c07\u5176\u65b7\u958b\uff0c\u800c\u7522\u751f\u300c\u7fa9\u7f8e\u300d \u3001 \u300c\u8c46\u5976\u300d \u3002 \u76ee\u6a19\u8a5e\u5f59\u7684\u5171\u73fe\u8a5e\u5f59\u548c\u5176\u540c\u7fa9\u8a5e\u7684\u5171\u73fe\u8a5e\u5f59\u61c9\u5177\u5099\u67d0\u7a2e\u7a0b\u5ea6\u7684\u76f8\u4f3c\u6027\uff0c\u4f8b\u5982\u300c\u7fa9\u7f8e \u8c46\u5976\u300d\u7684\u5171\u73fe\u8a5e\u5f59\u548c\u300c\u7fa9\u7f8e\u8c46\u6f3f\u300d\u7684\u5171\u73fe\u8a5e\u5f59\u61c9\u8a72\u985e\u4f3c\u3002\u56e0\u6b64\uff0c\u6211\u5011\u5f9e\u6240 PTT \u7db2\u8def\u8ad6 \u985e\u5225\u76f8\u540c\u7684\u7522\u54c1\u3002\u56e0\u6b64\uff0c\u6211\u5011\u9032\u4e00\u6b65\u63a1\u7528\u6df1\u5ea6\u5b78\u7fd2\u4e2d\u7684 Word2vec \u65b9\u6cd5\uff0c\u4ee5\u5f2d\u88dc\u5171\u73fe\u8a5e (\u4e94)\u3001 \u540c\u7fa9\u8a5e\u64f4\u589e \u5716\u4e8c\u3001Skip-gram \u6a21\u578b[1] \u53c3\u6578\u8a2d\u5b9a (https://code.google.com/archive/p/word2vec/)\u3002 \u5f8c\u7e8c\u7684\u540c\u7fa9\u8a5e\u8a5e\u5eab\u5efa\u7f6e\u3002 \u7684\u7f3a\u9ede\u3002 \u540c\u7fa9\u8a5e\u5f59\uff0c\u5728\u5be6\u9a57\u4e2d Word2vec \u6a21\u578b\u53c3\u6578\u5982\u8868\u4e8c\uff0c\u6b64\u53c3\u6578\u4f9d\u7167 Word2vec \u5b98\u65b9\u6587\u4ef6\u9032\u884c \u4e94\uff0c\u5176\u7be9\u9078\u76f8\u4f3c\u503c\u9580\u6abb\u70ba 0.5\uff0c\u900f\u904e\u672c\u65b9\u6cd5\u81ea\u52d5\u5316\u7be9\u9078\u904e\u5f8c\u4e4b\u540c\u7fa9\u8a5e\uff0c\u5c07\u63d0\u4f9b\u4eba\u5de5\u9032\u884c \u7d93\u904e\u524d\u9762\u5169\u500b\u6b65\u9a5f\u6240\u7522\u751f\u7684\u53ef\u80fd\u540c\u7fa9\u8a5e\uff0c\u8207\u5176\u6240\u5c0d\u61c9\u7684\u7522\u54c1\u540d\u7a31\u6bd4\u5c0d\u5f8c\u7684\u7d50\u679c\u5982\u8868 \u4ee3\u63db\u70ba\u300c\u53ef\u6a02\u300d \uff0c\u6b64\u4eba\u5de5\u8cc7\u6599\u505a\u70ba\u6e2c\u8a66\u8cc7\u6599\u96c6\uff0c\u5efa\u7acb\u76ee\u7684\u662f\u672c\u7814\u7a76\u662f\u5426\u80fd\u5920\u627e\u5230\u76ee\u6a19\u7684 (\u4e09)\u3001 \u5171\u73fe\u8a5e\u5f59\u96c6\u5efa\u7acb \u8a5e\u8b93\u5f8c\u7e8c\u7684\u7522\u54c1\u540d\u7a31\u6bd4\u5c0d\u6cd5\u9032\u884c\u6bd4\u5c0d\u3002 \u7136\u800c\uff0c\u6b64\u65b9\u6cd5\u4e26\u672a\u8003\u616e\u7522\u54c1\u8a5e\u5f59\u548c\u53e3\u7891\u8a5e\u5f59\u9593\u7684\u5b57\u5e8f\u95dc\u4fc2\uff0c\u56e0\u6b64\uff0c\u4e92\u76f8\u7af6\u722d\u7684\u7522\u54c1\uff0c \u5982\u7d71\u4e00\u548c\u7fa9\u7f8e\u7684\u8c46\u5976\uff0c\u5176\u6240\u4f7f\u7528\u7684\u53e3\u7891\u8a5e\u5f59\u6709\u53ef\u80fd\u5b8c\u5168\u76f8\u540c\uff0c\u56e0\u800c\u5728\u6b64\u968e\u6bb5\uff0c\u53ea\u80fd\u627e\u51fa \u7522\u54c1\u540d\u7a31 \u76f8\u4f3c\u503c \u7522\u54c1\u540d\u7a31 \u76f8\u4f3c\u503c \u7522\u54c1\u540d\u7a31 \u76f8\u4f3c\u503c \u4e4b\u53e3\u7891\u7684\u7522\u54c1\u540d\u7a31\u70ba\u7fa9\u7f8e\u8c46\u5976\u3001\u738b\u5b50\u9eb5\u3001\u53ef\u53e3\u53ef\u6a02\uff0c\u8cc7\u6599\u96c6\u4f86\u6e90\u70ba\u900f\u904e\u8cc7\u7b56\u6703\u7684 API(http://api.ser.ideas.iii.org.tw/)\u64f7\u53d6 PTT \u7684\u53e3\u7891\u8cc7\u6599\u3002\u4e26\u7531\u4eba\u70ba\u65b9\u5f0f\u7522\u751f\u540c\u6a23\u7b46\u6578\u4f46\u5c07 \u300c\u7fa9\u7f8e\u8c46\u5976\u300d\u7522\u54c1\u8a5e\u5f59\u4ee3\u63db\u70ba\u300c\u7fa9\u7f8e\u8c46\u6f3f\u300d \u3001 \u300c\u738b\u5b50\u9eb5\u300d\u4ee3\u63db\u70ba\u300c\u5c0f\u738b\u5b50\u9eb5\u300d \u3001 \u300c\u53ef\u53e3\u53ef\u6a02\u300d \u5c0f\u738b\u5b50\u9eb5 0.9982483 \u9ea5\u7d05 0.999598503 \u53ef\u6a02 0.997811 \u51ac\u7c89 0.9765478 \u670d\u52d9 0.995698630 \u592a\u53e4 \u70cf\u9f8d 0.9749084 \u4e3b\u5834 0.994993865 \u767e\u4e8b\u53ef\u6a02 0.985078 0.985727 \u53c3\u8003\u6587\u737b</td></tr><tr><td>\u4e00\u3001 \u524d\u8a00 \u672c\u7814\u7a76\u63d0\u51fa\u7522\u54c1\u540d\u7a31\u6a5f\u7387\u6bd4\u5c0d\u6cd5\u7d50\u5408\u8a9e\u610f\u6982\u5ff5\u6a21\u578b\u548cWord2vec\u7684\u65b9\u6cd5[1]\uff0c\u4ee5\u6a5f\u7387 \u65b9\u5f0f\u64f4\u589e\u7522\u54c1\u7684\u975e\u6b63\u898f\u7528\u8a5e\u70ba\u76ee\u6a19\u7522\u54c1\u7684\u540c\u7fa9\u8a5e\uff0c\u4ee5\u6539\u5584\u53e3\u7891\u6240\u63cf\u8ff0\u7684\u7522\u54c1\u540d\u7a31\u548c\u5ee0\u5546 \u7522\u54c1\u540d\u7a31\u4e0d\u4e00\u81f4\u7684\u554f\u984c\uff0c\u964d\u4f4e\u76ee\u6a19\u7522\u54c1\u53e3\u7891\u641c\u5c0b\u6642\uff0c\u6240\u7522\u751f\u7684\u8cc7\u8a0a\u907a\u6f0f\u3002 \u5b57\u3001\u540c\u97f3\u5b57\u3001\u65b0\u5275\u5b57\u3001\u932f\u5225\u5b57\u3001\u5225\u540d\u3001\u540c\u7fa9\u8a5e\u7b49\u975e\u6b63\u898f\u7528\u8a5e\u7684\u767c\u751f\u3002\u7576\u9019\u4e9b\u975e\u6b63\u898f\u7528\u8a5e\uff0c \u767c\u751f\u65bc\u7522\u54c1\u3001\u670d\u52d9\u6216\u516c\u53f8\u540d\u7a31\u6642\uff0c\u5c0e\u81f4\u63a1\u7528\u95dc\u9375\u5b57\u6536\u96c6\u53e3\u7891\u6642\uff0c\u56e0\u8207\u6b63\u5f0f\u540d\u7a31\u5b57\u578b\u4e0d\u540c\uff0c \u6703\u7522\u751f\u8cc7\u8a0a\u907a\u6f0f\u7684\u554f\u984c\uff0c\u4f8b\u5982\u6b32\u5c0b\u627e\u300c\u7fa9\u7f8e\u8c46\u5976\u300d\u7684\u53e3\u7891\u6587\u7ae0\uff0c\u5247\u6703\u907a\u6f0f\u8aa4\u6253\u70ba\u300c\u7fa9\u7f8e \u8c46\u6f3f\u300d\u7684\u53e3\u7891\u6587\u7ae0\u3002 \u5b9a\u7fa9\u7522\u54c1\u3001\u670d\u52d9\u6216\u516c\u53f8\u540d\u7a31\u7684\u6bd4\u5c0d\u7684\u7814\u7a76\u7a31\u70ba\u547d\u540d\u5be6\u9ad4\u8b58\u5225(NER; name entity recognition)\u3002\u5e38\u7528\u7684\u65b9\u6cd5\u6709\u4e8c\u7a2e\uff0c\u7b2c\u4e00\u7a2e\u70ba\u6cd5\u5247\u6cd5\uff0c\u7531\u4eba\u5de5\u5efa\u7acb\u6b63\u78ba\u7684\u547d\u540d\u5be6\u9ad4\u96c6\uff0c\u63a1 \u53d6\u6bd4\u5c0d\u65b9\u5f0f\uff1b\u7b2c\u4e8c\u7a2e\u70ba\u6a5f\u5668\u5b78\u7fd2\u6cd5\uff0c\u4f7f\u7528\u6a5f\u5668\u5b78\u7fd2\u6f14\u7b97\u6cd5\uff0c\u5f9e\u5df2\u6a19\u8a18\u7684\u8cc7\u6599\u96c6\u5efa\u7acb\u547d\u540d \u5be6\u9ad4\u8b58\u5225\u6a21\u578b[6]\u3002\u6587\u737b\u4e0a\uff0c\u547d\u540d\u5be6\u9ad4\u7814\u7a76\uff0c\u4ee5\u627e\u51fa\u6587\u7ae0\u4e2d\u4e8b\u5148\u5b9a\u7fa9\u7684\u547d\u540d\u5be6\u9ad4\u985e\u5225\u70ba \u4e3b\u8981\u7684\u76ee\u7684\uff0c\u4f8b\u5982\u627e\u51fa\u4eba\u540d\u3001\u5730\u540d\u6216\u662f\u516c\u53f8\u540d\u7684\u5be6\u4f8b[7]\uff0c\u4e26\u5ffd\u7565\u56e0\u7e2e\u5beb\u5b57\u3001\u540c\u97f3\u5b57\u3001 \u65b0\u5275\u5b57\u3001\u932f\u5225\u5b57\u3001\u5225\u540d\u6240\u7522\u751f\u7684\u540c\u7fa9\u8a5e\u7b49\u975e\u6b63\u898f\u7528\u8a5e\u7684\u5b58\u5728\u554f\u984c\u3002\u4ea6\u5373\uff0c\u50b3\u7d71NER\u7684\u7814 \u7a76\uff0c\u53ef\u4ee5\u5224\u65b7\u547d\u540d\u5be6\u9ad4A\u548cB\u540c\u5c6c\u65bc\u4eba\u540d\u3001\u5730\u540d\u6216\u662f\u516c\u53f8\u540d\uff0c\u4f46\u4e26\u4e0d\u8655\u7406\u547d\u540d\u5be6\u9ad4A\u548c \u7a2e\u985e\u4e5f\u662f\u76f8\u7576\u591a\u5143\u7684\uff0c\u76ee\u524d\u4e3b\u6d41\u7684\u7814\u7a76\u90fd\u662f\u4ee5\u82f1\u6587\u70ba\u4e3b[11]\uff0c\u4e5f\u6709\u4ee5\u571f\u8033\u5176\u8a9e\u70ba\u57fa\u790e\u7684 \u6df7\u5408\u578b\u547d\u540d\u5be6\u9ad4\u8b58\u5225\uff0c\u5176\u900f\u904e\u898f\u5247\u7684\u5efa\u7acb\u9032\u884c\u8b58\u5225\uff0c\u6b64\u6a21\u578b\u53ef\u4ee5\u540c\u6642\u5c0d\u65bc\u65b0\u805e\u3001\u8ca1\u7d93\u65b0 \u805e\u3001\u7ae5\u8a71\u6545\u4e8b\u3001\u6b77\u53f2\u6587\u5b57\u8cc7\u6599\u9032\u884c\u547d\u540d\u5be6\u9ad4\u7684\u8b58\u5225\uff0c\u5176\u6240\u63d0\u51fa\u7684\u6a21\u578b\u6709\u826f\u597d\u7684\u8b58\u5225\u7cbe\u78ba \u7387\uff0c\u7f3a\u9ede\u5247\u70ba\u5fc5\u9808\u4f9d\u8cf4\u7e41\u7463\u7684\u898f\u5247[11]\u3002 (\u4e8c)\u3001 \u6587\u5b57\u5411\u91cf\u8868\u793a \u5411\u91cf\u7a7a\u9593\u6a21\u578b(VSM; vector space model)\u70ba\u4e3b\u8981\u7684\u6587\u5b57\u5411\u91cf\u8868\u793a\u65b9\u6cd5\uff0c\u5ee3\u6cdb\u904b\u7528\u65bc\u6587 \u5b57\u63a2\u52d8\u3001\u8cc7\u8a0a\u6aa2\u7d22\u3001\u81ea\u7136\u8a9e\u8a00\u8655\u7406\u7b49\u7814\u7a76\u9818\u57df\u4e2d\u3002\u5411\u91cf\u7a7a\u9593\u6a21\u578b\u5c07\u5b57\u8996\u70ba\u5411\u91cf\u7684\u7d14\u91cf\u6216 \u5c6c\u6027\uff0c\u8cc7\u6599\u96c6\u4e2d\u7684\u6240\u6709\u4e0d\u540c\u5b57\u69cb\u6210\u9ad8\u7dad\u5ea6\u5411\u91cf\u7a7a\u9593\uff0c\u4e00\u7bc7\u6587\u7ae0\u4f7f\u7528\u4e00\u500b\u6587\u5b57\u5411\u91cf\u4f86\u8868 \u793a\uff0c\u5169\u7bc7\u6982\u5ff5\u985e\u4f3c\u7684\u6587\u7ae0\u56e0\u4f7f\u7528\u985e\u4f3c\u7684\u6587\u5b57\uff0c\u9810\u671f\u6620\u5c04\u5230\u76f8\u8fd1\u7684\u5411\u91cf\u7a7a\u9593\u3002\u55ae\u7d14\u7684\u6587\u5b57 \u4ea6\u53ef\u5411\u91cf\u5316\uff0c\u76f8\u4f3c\u7684\u6587\u5b57\u5c07\u51fa\u73fe\u65bc\u6982\u5ff5\u985e\u4f3c\u7684\u6587\u7ae0\u4e2d\uff0c\u56e0\u6b64\uff0c\u6982\u5ff5\u985e\u4f3c\u7684\u6587\u5b57\uff0c\u9810\u671f\u80fd \u6620\u5c04\u7684\u76f8\u8fd1\u7684\u5411\u91cf\u7a7a\u9593\u4e2d\u3002Baroni [12]\u5c07\u5206\u6790\u6587\u5b57\u5411\u91cf\u9593\u7684\u95dc\u4fc2\u5340\u5206\u70ba\u5169\u7a2e\u65b9\u6cd5\uff1a\u8a5e\u983b \u7684\u6587\u7ae0\u6240\u958b\u767c\u3002Word2vec \u4e3b\u8981\u7684\u529f\u80fd\u70ba\u5c07\u8fad\u5f59\u8f49\u5316\u6210\u70ba\u6587\u5b57\u5411\u91cf\uff0c\u900f\u904e\u5411\u91cf\u7a7a\u9593\u6a21\u578b \u53ef\u4ee5\u8a08\u7b97\u8a9e\u7fa9\u9593\u7684\u76f8\u4f3c\u5ea6\uff0c\u5c6c\u65bc\u985e\u795e\u7d93\u6a5f\u7387\u8a9e\u8a00\u6a21\u578b\u7684\u4e00\u7a2e\u3002\u5176\u6a21\u578b\u7684\u9810\u6e2c\u65b9\u5f0f\u5206\u70ba\u5169 \u7a2e\uff0c\u5206\u5225\u662f CBOW (continuous bag-of-word)\u548c Skip-Gram [1]\u3002CBOW \u4e26\u672a\u4f7f\u7528\u985e\u795e\u7d93 \u7db2\u8def\u5e38\u7528\u7684\u975e\u7dda\u6027\u96b1\u85cf\u5c64(non-linear hidden layer)\uff0c\u5728\u8f38\u5165\u5c64\u7684\u6240\u6709\u55ae\u8a5e\u7686\u5171\u4eab\u96b1\u85cf\u5c64\uff0c \u5176\u8a13\u7df4\u76ee\u6a19\u662f\u7d66\u5b9a\u4e00\u500b\u76ee\u6a19\u8a5e\u7684\u4e0a\u4e0b\u6587\u9130\u8fd1\u8a5e\uff0c\u4ee5\u9810\u6e2c\u76ee\u6a19\u8a5e\u51fa\u73fe\u7684\u6a5f\u7387\uff0c\u6b64\u65b9\u6cd5\u9069\u5408 \u8f03\u5c0f\u7684\u8cc7\u6599\u96c6\u3002Skip-gram \u8207 CBOW \u4e0d\u540c\uff0c\u4f7f\u7528\u4e00\u4e32\u6587\u5b57\u4e2d\u7684\u4e00\u500b\u76ee\u6a19\u8a5e\uff0c\u4f86\u9810\u6e2c\u9130\u8fd1 \u8a5e\u767c\u751f\u7684\u6a5f\u7387\uff0c\u6b64\u65b9\u6cd5\u9069\u5408\u8f03\u5927\u7684\u8cc7\u6599\u96c6\u3002Word2vec \u4e4b\u6240\u4ee5\u6703\u53d7\u5230\u95dc\u6ce8\u662f\u56e0\u70ba Word2vec \u7684\u9ad8\u6548\u7387\u548c\u53ef\u7528\u6027\uff0c\u56e0\u70ba Word2vec \u4e0d\u50cf\u985e\u795e\u7d93\u7db2\u8def\u67b6\u69cb\u7684\u65b9\u5f0f\uff0c\u5fc5\u9808\u4f7f\u7528\u5927\u91cf\u7684\u8a13\u7df4 \u8a5e\u5f59\u5411\u91cf\uff0c\u5373\u53ef\u9810\u6e2c\u4e0d\u540c\u76ee\u6a19\u8a5e\u7684\u9130\u8fd1\u8a5e\u51fa\u73fe\u7684\u6a5f\u7387[15]\u3002Word2vec \u9069\u5408\u505a\u70ba\u6587\u5b57\u5411\u91cf \u7279\u5fb5\u7684\u904b\u7b97\u5de5\u5177\uff0c\u5982 Zhang [16]\u900f\u904e Word2vec \u9032\u884c\u8a9e\u610f\u7279\u5fb5\u7684\u8a08\u7b97\u518d\u5229\u7528 SVM \u5206\u985e\u5668 \u9032\u884c\u6587\u672c\u7684\u60c5\u611f\u5206\u985e\uff0c\u7d93\u5be6\u9a57\u5f97\u77e5\uff0c\u6b64\u65b9\u6cd5\u80fd\u5920\u5f97\u5230\u76f8\u7576\u9ad8\u7684\u5206\u985e\u6b63\u78ba\u7387\u3002 \u5716\u4e00\u3001\u7522\u54c1\u540d\u7a31\u6a5f\u7387\u6bd4\u5c0d\u6cd5 (\u4e00)\u3001 \u53e3\u7891\u6536\u96c6 \u91dd\u5c0d\u4e2d\u6587\u60c5\u611f\u53e3\u7891\u7db2\u7ad9\uff0c\u5982 PTT \u7db2\u8def\u8ad6\u58c7(http://ptt.cc)\u8490\u96c6\u53e3\u7891\u8cc7\u6599\uff0c\u4e26\u53bb\u9664 HTML \u6a19\u7c64\u53ca\u975e\u672c\u6587\u5167\u5bb9\uff0c\u63d0\u53d6\u53e3\u7891\u6587\u672c\u3002\u672c\u7814\u7a76\u53e3\u7891\u6587\u7ae0\u7684\u6536\u96c6\u4f9d\u8cf4\u95dc\u9375\u5b57\uff0c\u56e0\u6b64\u6240\u6536\u96c6\u7684 \u53e3\u7891\u6587\u7ae0\uff0c\u5747\u542b\u6709\u76ee\u6a19\u95dc\u9375\u5b57\uff0c\u5728\u521d\u6b65\u7684\u7814\u7a76\u4e2d\uff0c\u672c\u7814\u7a76\u63a1\u53d6\u6a21\u64ec\u65b9\u5f0f\uff0c\u5c07\u6240\u6536\u96c6\u7684\u53e3 \u7891\u6587\u7ae0\u6dfb\u52a0\u76ee\u6a19\u95dc\u9375\u5b57\u7684\u540c\u7fa9\u8a5e\uff0c\u5982\u8907\u88fd\u6240\u6536\u96c6\u6709\u95dc\u300c\u7fa9\u7f8e\u8c46\u5976\u300d\u7684\u53e3\u7891\u6587\u7ae0\uff0c\u4f46\u5c07\u5176 \u76ee\u6a19\u95dc\u9375\u5b57\uff0c\u5982\u300c\u7fa9\u7f8e\u8c46\u5976\u300d\u4fee\u6539\u70ba\u300c\u7fa9\u7f8e\u8c46\u6f3f\u300d \u3002 (\u4e8c)\u3001 \u8cc7\u6599\u524d\u8655\u7406 \u4f9d\u524d\u6b65\u9a5f\u6240\u6536\u96c6\u7684\u53e3\u7891\u96d6\u5df2\u7121 HTML \u6a19\u7c64\u3001CSS \u6a19\u7c64\u3001Java script \u8a9e\u6cd5\uff0c\u4f46\u4ecd\u542b\u6709 \u90e8\u5206\u96dc\u8a0a\uff0c\u5982\u6a19\u9ede\u7b26\u865f\u6216\u662f\u7279\u6b8a\u7684\u6a19\u7c64\uff0c\u672c\u7814\u7a76\u53bb\u9664\u975e\u4e2d\u6587\u5b57\u5143\uff0c\u63a1\u7528 Jieba (Jieba \u8868\u4e00\u3001\u5171\u73fe\u8a5e\u5f59\u96c6 \u967d\u5149 \u7fa9\u7f8e\u8c46\u5976 \u597d\u559d \u7121\u7cd6 \u967d\u5149 -1 1 1 \u7fa9\u7f8e\u8c46\u5976 1 -1 2 \u597d\u559d 1 1 -1 \u7121\u7cd6 1 2 1 \u7684\u5b57\u5e8f\u95dc\u4fc2\u3002\u7576\u300c\u7fa9\u7f8e\u8c46\u5976\u300d\u548c\u300c\u7fa9\u7f8e\u8c46\u6f3f\u300d\u6240\u7522\u751f\u7684\u5171\u73fe\u8a5e\u5177\u5099\u9ad8\u5ea6\u7684\u91cd\u8907\u6027\u6642\uff0c\u5247 \u53ef\u63a8\u65b7\u70ba\u540c\u7fa9\u8a5e\u3002 Word2vec \u8a13\u7df4\u6642\uff0c\u5fc5\u9808\u8f38\u5165\u6240\u8981\u8003\u616e\u7684\u5b57\u5e8f\u9577\u5ea6\uff0c\u7576\u8003\u616e\u7684\u5b57\u5e8f\u6108\u9577\u6642\uff0c\u6240\u9700\u8a13 \u7df4\u7684\u6642\u9593\u6108\u4e45\uff0c\u82e5\u8a2d\u5b9a\u7684\u5b57\u5e8f\u9577\u5ea6\u70ba 2\uff0c\u4ea6\u5373\u8003\u616e\u76ee\u6a19\u8a5e(\u5982\u7fa9\u7f8e\u8c46\u5976)\u524d 1~2 \u500b\u5b57\u548c\u5f8c \u7684\u7522\u54c1\u540d\u7a31\uff0c\u8a08\u7b97\u76f8\u4f3c\u503c\uff0c\u7576\u7b26\u5408\u9580\u6abb\u503c\u6642\uff0c\u4ee5\u76f8\u4f3c\u5ea6\u6700\u9ad8\u8005\u70ba\u5176\u7522\u54c1\u540d\u7a31\uff0c\u4e26\u5c07\u76f8\u4f3c (\u4e8c)\u3001 \u5be6\u9a57\u7d50\u679c \u4e94\u3001 \u7d50\u8ad6\u8207\u672a\u4f86\u5c55\u671b \u7684\u7522\u54c1\u540d\u7a31\u5217\u5165\u7522\u54c1\u8a9e\u6599\u5eab\u4e2d\uff0c\u76f8\u4f3c\u5ea6\u7684\u8a08\u7b97\u65b9\u5f0f\u63a1\u53d6\u53e3\u7891\u6587\u7ae0\u4e2d\u7522\u54c1\u55ae\u4e00\u4e2d\u6587\u5b57 (Chinese character)\u4f54\u6240\u6bd4\u5c0d\u7684\u53e3\u7891\u8a9e\u6599\u5eab\u7522\u54c1\u540d\u7a31\u7e3d\u5b57\u6578\u767e\u5206\u6bd4\u3002\u76f8\u4f3c\u5ea6\u516c\u5f0f\u5982\u516c\u5f0f(2) \u6240\u793a\uff1a \u900f\u904e 2141 \u7b46\u53e3\u7891\u8cc7\u6599\u6240\u8a13\u7df4\u7684\u7fa9\u7f8e\u8c46\u5976\u3001\u738b\u5b50\u9eb5\u3001\u53ef\u53e3\u53ef\u6a02\u7684\u5171\u73fe\u8a5e\u5f59\u96c6\uff0c\u8207\u4eba \u672c\u7814\u7a76\u6240\u63d0\u51fa\u7684\u7522\u54c1\u6a5f\u7387\u6bd4\u5c0d\u6cd5\uff0c\u4e3b\u8981\u5229\u7528\u7522\u54c1\u8a5e\u5f59\u8207\u53e3\u7891\u8a5e\u5f59\u7684\u5171\u73fe\u95dc\u4fc2\uff0c\u627e\u51fa \u5de5\u5efa\u7acb\u4e4b\u6e2c\u8a66\u53e3\u7891\u8cc7\u6599\u7d93\u904e\u6bd4\u5c0d\u3001\u6bd4\u5c0d\u9580\u6abb\u70ba 50%\uff0c\u8207\u7fa9\u7f8e\u8c46\u5976\u53ef\u80fd\u70ba\u540c\u7fa9\u8a5e\u7684\u8a5e\u5f59 \u548c\u76ee\u6a19\u7522\u54c1\u5177\u5099\u9ad8\u5ea6\u95dc\u4fc2\u7684\u540c\u7fa9\u8a5e\uff0c\u63a5\u8457\u4f7f\u7528 Word2vec \u6a21\u578b\u9032\u884c\u540c\u7fa9\u8a5e\u64f4\u589e\u3002\u6839\u64da\u672c \u70ba\uff0c\u7d71\u4e00\u8c46\u6f3f\u3001\u5168\u806f\u3001\u71d5\u9ea5\u3001\u7fa9\u7f8e\u8c46\u6f3f\u2026\u7b49\u7b49\u3002\u900f\u904e\u5171\u73fe\u8a5e\u6bd4\u5c0d\u6240\u7522\u751f\u7684\u8a5e\u5f59\u773e\u591a\u5728\u6b64 -Cos(a,b)= (1) \u5176\u4e2d a \u8868\u53e3\u7891\u6587\u7ae0\u4e2d\u7684\u7522\u54c1\u8a5e\u5f59\uff0cb \u8868\u53e3\u7891\u6587\u7ae0\u4e2d\u7684\u53e3\u7891\u8a5e\u5f59\u3002 (\u56db)\u3001 \u76f8\u4f3c\u5ea6\u6bd4\u5c0d \u7576\u8f38\u5165\u672a\u77e5\u7522\u54c1\u7684\u53e3\u7891\u6587\u7ae0\u5f8c\uff0c\u5982\u8a72\u6587\u7ae0\u5305\u542b \u300c\u7fa9\u7f8e\u8c46\u6f3f\u300d \u4f46\u4e26\u672a\u5305\u542b \u300c\u7fa9\u7f8e\u8c46\u5976\u300d \uff0c \u5c0d\u65bc\u672a\u77e5\u7522\u54c1\u7684\u53e3\u7891\u6587\u7ae0\uff0c\u540c\u6a23\u7d93\u904e\u8cc7\u6599\u524d\u8655\u7406\u6b65\u9a5f\u5f8c\uff0c\u5176\u53e3\u7891\u8a5e\u5f59\u9ad8\u65bc\u5171\u73fe\u8a5e\u5f59\u96c6\u4e00 1~2 \u500b\u5b57\u7684\u95dc\u4fc2\u3002\u8209\u4f8b\u5982\u4e0b\uff1a (1)\u5168\u5bb6 \u6709 \u7fa9\u7f8e\u8c46\u5976 \u7121\u7cd6 \u597d\u559d (2)\u5168\u5bb6 \u6709 \u7fa9\u7f8e\u8c46\u6f3f \u7121\u7cd6 \u597d\u559d \u6b64\u7bc4\u4f8b\u4e2d\u5169\u53e5\u4e2d\u53ea\u6709\u300c\u7fa9\u7f8e\u8c46\u5976\u300d\u8207\u300c\u7fa9\u7f8e\u8c46\u6f3f\u300d\u4e0d\u540c\uff0c\u6574\u53e5\u8a71\u6240\u63cf\u8ff0\u7684\u4e3b\u89d2\u5176\u5be6 \u6700\u5927\u76f8\u4f3c\u5ea6= (2) \u7814\u7a76\u7684\u521d\u6b65\u5be6\u4f5c\u7d50\u679c\uff0c\u767c\u73fe\u672c\u7814\u7a76\u6240\u63d0\u51fa\u7684\u65b9\u6cd5\uff0c\u5177\u5099\u9ad8\u5ea6\u7684\u61c9\u7528\u6f5b\u529b\u3002\u6587\u737b\u4e0a\uff0c\u5728\u547d \u4e0d\u4e00\u4e00\u5217\u51fa\u3002\u6839\u64da Word2vec \u6a21\u578b\u53c3\u6578\u6240\u8a08\u7b97\u51fa\u8207\u300c\u7fa9\u7f8e\u8c46\u5976\u300d\u76f8\u4f3c\u7684\u8a5e\u5f59\u5982\u8868\u4e09\u6240\u793a\uff0c \u540d\u5be6\u9ad4\u8b58\u5225(NER)\u7684\u7814\u7a76\u4e2d\uff0c\u5f88\u5c11\u88ab\u61c9\u7528\u65bc\u540c\u7fa9\u8a5e\u7684\u8fa8\u8b58\uff0c\u672c\u7814\u7a76\u7684\u63d0\u51fa\uff0c\u9664\u4e86\u80fd\u904b\u7528 \u7814\u7a76\u5c07\u8a13\u7df4\u5b8c\u6210\u6a21\u578b\uff0c\u900f\u904e Python \u7684 Gensim \u5957\u4ef6\u8a08\u7b97\u8a5e\u5f59\u7684\u76f8\u4f3c\u5ea6\uff0c\u5176\u986f\u793a\u6700\u76f8\u4f3c \u8a5e\u6392\u5e8f\u5206\u5225\u70ba\u7fa9\u7f8e\u8c46\u6f3f\u3001\u7fa9\u7f8e\u3001\u98df\u54c1\u3002\u5728\u6b64\u6703\u5c07\u53ef\u80fd\u7684\u540c\u7fa9\u8a5e\u9032\u884c\u5f8c\u7e8c\u7684\u7522\u54c1\u540d\u7a31\u6bd4\u5c0d\u3002 \u65bc\u53e3\u7891\u641c\u5c0b\u4e2d\uff0c\u6e1b\u5c11\u8cc7\u8a0a\u907a\u6f0f\u7684\u554f\u984c\u5916\uff0c\u4e5f\u5c0d\u65bc NER \u7814\u7a76\uff0c\u64f4\u5145\u540c\u7fa9\u8a5e\u7814\u7a76\u7684\u65b9\u5411\u3002 \u5176\u4e2d w i \u5b57\u6578\u3002 \u672a\u4f86\u7684\u767c\u5c55\u53ef\u4ee5\u66f4\u9032\u4e00\u6b65\u63a1\u7528\u6a5f\u7387\u6a21\u578b\u5982\u4ea4\u4e92\u8cc7\u8a0a(MI; mutual information)\u6a21\u578b\uff0c \u8868\u4e09\u3001\u7fa9\u7f8e\u8c46\u5976\u76f8\u4f3c\u8a5e\u5f59\u8868 \u7522\u54c1\u540d\u7a31 \u76f8\u4f3c\u503c \u6539\u5584\u5171\u73fe\u8a5e\u5f59\u96c6\uff0c\u53e6\u5916\u4e5f\u53ef\u904b\u7528\u4e3b\u984c\u6a21\u578b(topic model)\uff0c\u627e\u51fa\u66f4\u5177\u4ee3\u8868\u6027\u7684\u8a5e\u5f59\uff0c\u8b93\u5411 \u90fd\u662f\u300c\u7fa9\u7f8e\u8c46\u5976\u300d\u9019\u4e00\u500b\u7522\u54c1\u3002\u9810\u671f Word2vec \u80fd\u5920\u85c9\u8457\u5b57\u5e8f\u95dc\u4fc2\u627e\u51fa\u8f03\u5177\u53ef\u80fd\u7684\u540c\u7fa9 \u8a5e\u5f59\u3002 \u7fa9\u7f8e \u98df\u54c1 0.999582290649 0.999646663666 \u7fa9\u7f8e\u8c46\u6f3f 0.999682784081 \u91cf\u7a7a\u9593\u6a21\u578b\u66f4\u52a0\u7dca\u5bc6\u3002</td></tr></table>" |
| } |
| } |
| } |
| } |
