產品評論的正負面使用效果擷取之研究

Abstract

近來由於大眾對皮膚保養觀念的普及，加上網路平台資訊交流的便利性，許多消費者會透過網路取得商品資訊以及評論做為購買考量，由於評論內容多為個人使用心得的寫作結構，所以消費者往往需要花費許多時間精力閱讀以取得有用資訊，因此如能快速整理網路上的保養品評論，可以幫助消費者更快過濾有用資訊外，並提供保養品廠商有用的產品使用報告。 消費者觀看保養品評論，最主要希望得到產品使用效果以及評價。本研究透過斷詞與詞性標記工具，以評論內容的個別句子做為分析單位，計算兩兩字詞的關聯程度進行六種保養品使用效果分群，分析PMI值建立正負面詞庫，提出關鍵字詞擷取規則以及正負面語意判斷方法，可以有效判斷與使用效果有關的句子所表達的正面或負面使用經驗。 實驗結果證實本研究所提出的正負面產品使用效果判斷方法能夠有效的判斷保養品評論的正負面產品效果。

As the skin care concept is becoming popular and with the convenience of information exchange through Internet, more consumers could get product reviews of product information and use experience from the web for purchasing references. Because of diverse personal writing styles, consumers have to spend a lot of reading time to retrieve useful information. If these reviews can be organized quickly into keywords, it will be much helpful for consumers to filter information and can be used as usage summary report for skin care vendors. When consumers read the product reviews, their major purpose usually is to get the product effect and consumer evaluations. This study proposes a positive/negative product effect extraction methodology to effectively identify the positive/negative product effect from product reviews. The major steps of the proposed methodology includes: 1. Defining rules of skin care product review keyword retrieval and rules of positive/negative opinion identification. 2. Leveraging word segmentation and part-of-speech tagging tool, and uses the review sentence as the analysis unit to calculate connectivity between keywords and group them into six product effect clusters. 3. Analyzing PMI calculation result to build positive/negative keyword dictionary. The experiment result shows that the proposed methodology can effectively determine the positive/negative product effects of skin care product reviews.