新聞文件中意見句自動擷取及意見持有者辨識之研究

Automatically Extracting Opinion Sentences and Identifying Opinion Holders in News

陳崇儒(Chen Choung-Ru)

指導教授 ： 侯文娟

國立臺灣師範大學/理學院/資訊工程學系/碩士(2018年)

https://doi.org/10.6345/THE.NTNU.DCSIE.002.2018.B02

摘要

網路的發達，帶給人們便利。但每天都有大量的文本資訊需要閱讀，這時便可利用意見探勘擷取文本中人們感興趣之部分。而通常人們對文章會感興趣的部分都是誰發表什麼意見或是誰提出什麼看法，而這些描述的句子在文章中便稱為意見句。本研究提出監督式之機器學習方法，首先找出文章的意見句，再辨識意見句中的文章作者意見以及意見持有者。
利用自然語言處理之方法辨識文章作者以及意見持有者，其中方法包括Tokenization、蒐集意見詞、Stemming、尋找意見句、詞性標記、具名實體辨識和文章作者以及意見持有者之特徵擷取。而在特徵擷取部分，本論文利用詞彙相關資訊、詞性相關資訊、標點符號相關資訊、具名實體相關資訊、句法相關資訊、意見詞資訊以及文句組成相關資訊等特徵辨識文章中意見句之文章作者意見以及意見持有者。
實驗成果顯示在英語新聞文章中，文章作者意見辨識可以達到F-1值69.05%的效能；意見持有者辨識可以達到F-1值72.06%的效能。
關鍵字：意見探勘、意見句擷取、意見持有者辨識、機器學習、監督式學習

關鍵字

意見探勘 ； 意見句擷取 ； 意見持有者辨識 ； 機器學習 ； 監督式學習

並列摘要

Network of development gives people some convenience. However, there is a great deal of textual information that we need to read every day, so that we can utilize the opinion exploration to capture the part of the text we are interested in. Usually, people interested in who made comments or opinions in the article, and which are called opinion holders. This study proposes a supervised machine learning method. First we find the opinion of the article, and then identify the author of the article in the opinion and the holder of the opinion.
The method of natural language processing is used to identify the author of the article as well as the opinion holder, in which the method includes tokenization, collecting opinion words, stemming, finding opinion, part-of-speech tagging, recognizing the named entity and the author of the article and the feature extraction. In the feature extraction section, thesis dissertation uses the features of lexical related information, part of speech related information, punctuation related information, named entity related information, syntactic related information, opinion word information and sentence information to identify the article's opinion sentences, author's opinions and opinions holder.
The experimental results show that, the article author's opinion recognition can achieve 69.05% of the F-1 value and the opinion holder extraction can get 72.06% of the F-1 value.
Keywords: opinion exploration, opinion extraction, opinion holder identification, machine learning, supervised learning

並列關鍵字

opinion exploration ； opinion extraction ； opinion holder identification ； machine learning ； supervised learning

參考文獻

[1] Chang, C. C., & Lin, C. J. (2011). "LIBSVM: a library for support vector machines." ACM Transactions on Intelligent Systems and Technology (TIST), 2(3), 27.

[2] Choi, Y., Cardie, C., Riloff, E., & Patwardhan, S. (2005, October). "Identifying sources of opinions with conditional random fields and extraction patterns." In Proceedings of the conference on Human Language Technology and Empirical Methods in Natural Language Processing (pp. 355-362).

[3] Cortes, C., & Vapnik, V. (1995). "Support-vector networks." Machine learning, 20(3), 273-297.

[4] EBS美樂顧問中心-英語實力養成、專業能力發揮: http://ebseducation. pixnet.net/blog/post/406370224%E3%80%90%E5%95%86%E7%94%A8email%E3%80%919%E5%80%8B%E5%B8%B8%E7%94%A8%E7%9A%84%E5%8B%95%E8%A9%9E,%E6%9C%89%E6%95%88%E7%8E%87%E7%A2%BA%E5%AF%A6%E8%A1%A8%E9%81%94acknowledge

[5] Exam English Home Page: https://www.examenglish.com/vocabulary/b1_perso nal_feelings.htm