電影評論之助益性分析研究

Abstract

現今網際網路的蓬勃發展下，巨大的資料量已經是無可避免的趨勢，其中也包含了使用者留下的評論。眾多的評論中不一定每則都是有用的資訊，因此從大量的使用者評論中篩選出有助益性的評論，是本篇論文的研究目標。 評論的有助益性(review helpfulness)並沒有一個標準的定義，只要能幫助使用者有所思考，就能是助益性的一種。因此本研究嘗試透過各項特徵給定評論分數，作為判斷的依據。 本篇論文以雅虎電影中文短篇評論做為研究題材，使用中央研究院中文斷詞系統先將評論進行斷詞處理，再從資料裡找出TFIDF關鍵詞、詞性及評論長度。其中TFIDF關鍵詞經過教育部線上辭典進行同/反義詞擴充，並使用臺灣大學建立的情緒詞詞典NTUSD (National Taiwan University Semantic Dictionary)進行比對，找出每則評論所包含的情緒詞，且計算情緒詞出現的次數。並使用SVM訓練模型及預測結果，得到了79.7%的準確率。

With the rapid development of the Internet, huge amount of information is an inevitable trend, which also includes lots of user comments. Many reviews do not include useful information, so extracting helpful comments from a large number of user reviews is the research goal of this paper. There is no standard definition of review helpfulness, and as long as if it helps users to think about it, it can be helpful. Therefore, this study attempts to give comments by the characteristics of scores , as a basis for judgment. This thesis takes the short stories of Yahoo movie as the research target.The study uses the CKIP (Chinese Knowledge Information Processing) to process the comments first, and then find out the TFIDF keywords, parts of speech and lengths of comments from the data. The TFIDF keyword are expanded to synonyms and antonyms by the online dictionary of Ministry of Education. NTUSD (National Taiwan University Semantic Dictionary) was used built by National Taiwan University to find out the sentiment words contained in each comment and to calculate the number of sentiment words. Using SVM training model and prediction results, the accuracy of 79.7% was obtained.