運用情境分析與分解機之評論分析與評分預測

Abstract

線上評論網站是方便取得資訊的經驗分享平台，提供使用者參考評論意見與評分，並輔助使用者完成消費行為。然而由於大量的評論資訊造成資訊過載的問題，使用者不容易找尋符合自己喜好的項目，儘管線上評論網站大多都建立了推薦系統，但傳統方法只參考評分分數而忽略評論中的文字內容，導致使用者對於被推薦項目的期望產生落差。因此，分析線上評論網站之評論內容與評分來預測使用者之喜好評分，並提升推薦品質為重要之研究議題。 本研究提出基於使用者喜好面向與情境資訊之新評分預測方法，考量不同面向的評分因素，包括使用者對於不同面向的喜好重視程度以及評論之情境資訊。本研究中實驗運用文字探勘來發掘評論文字中各面向的意見特徵，並以此為基礎分析使用者喜好之面向，並加上使用者與項目的情境資訊進而預測使用者對不同項目之評分。本研究使用線上評論網站Yelp之資料實驗論文方法，結果顯示，本研究所提的方法優於傳統評分預測方法，能改善評分預測的準確性。

Online review website is a popular experience-sharing platform that provides users with reference items and helps users to choose products. However, due to the information overload problem, it is not easy for users to find items that meet their preferences. Moreover, the traditional recommendation approaches usually consider users ratings and ignore the review contents. Those recommended results performed by the traditional approaches could not fit for users’ real preferences. Therefore, it is important to analyze the review contents and ratings of reviews to predict the user's preference rating and improve the recommended quality as an important research topic. This study proposes a new rating prediction method based on user preferences and contextual information. The method takes into account the different rating factors which include the users’ preferences on different aspects and the contexts of the reviews. In this study, we use text mining as a basis to analyze user preferences, and add users and items contextual information into prediction models to improve the prediction results. The results show that the method proposed in this study outperforms the traditional methods and improves the accuracy of rating prediction.