侯文娟陳昱年2019-09-052015-8-72019-09-052013http://etds.lib.ntnu.edu.tw/cgi-bin/gs32/gsweb.cgi?o=dstdcdr&s=id=%22GN060047020S%22.&%22.id.&http://rportal.lib.ntnu.edu.tw:80/handle/20.500.12235/106542目前情緒語意分析的研究，半監督式學習和非監督式學習還屬於初期發展的階段。由於監督式學習情緒分析的研究已經相對於成熟很多，所以非監督式學習的情緒分析將會是未來的研究目標。 過去在詞彙極性分類方面多數人都是採用人工標註的方法，雖然準確度高但是所耗費的時間人力卻是一大問題，而在專業領域中詞彙往往又有不同的意義，也使得人工標註難度更加提高。 本篇論文使用網路論壇上的長篇中文電影評論集，探討中文文章中情緒詞彙的分類，盡可能的把所有在電影領域中可能帶有情緒意義的詞彙進行二元化『正向極性』以及『負向極性』的分類。 在此以非監督式方法進行，過程中不需要人工的介入，使用自定義的語法規則找出『種子詞彙』，接著利用教育部提供的詞典進行『同義詞』和『反義詞』的擴充詞彙，最後使用『模糊比對』等的步驟進行極性分類。The semi-supervised learning and unsupervised learning are both at the beginning of development in sentiment analysis. Since the supervised learning of sentiment analysis has been well-developed, the unsupervised learning is the objective in further research. Most people manually made polarity classification in the vocabulary in the past. It can reach high accuracy but the problem rises in consuming much time and human efforts. Furthermore, words often have differentmeanings in various professional field, it also makes more difficult in improving the manual annotation. In this thesis, we use Chinese movie reviews with a large number of words in network forum to explore the Chinese polarity classification. We classify the emotional words into two categories "positive polarity" and "negative polarity". We use unsupervised methods in this thesis. The process of the unsupervised method does not require manual intervention. Then we propose the syntactic rules to identify "seed words". In the following, we use the "synonym set" and "antonym set" from the dictionary provided by the Ministry of Education to expand the "seed words". The "fuzzy match" steps to classify polarity are applied in the final.自然語言處理語意分類非監督式學習中文處理