- 學位論文
基於情緒分析之評論等級預測
THE COMMENT LEVEL PREDICTION BASED ON SENTIMENT ANALYSIS
摘要
現今消費者愈來愈傾向在網路上進行購物及對所消費之店家進行評論，透過情緒分析，消費者的意見可以被截取出來，而這些評論會影響其他想要消費的消費者之意願。對於服務或商品的評論，有些是文字評論，有些是給予星號等級，有些是二者皆有。 雖然文字評論能給予較詳細的說明，但是面對大量的評論，消費者無法每一則評論都瀏覽過，所以星號等級就成了消費者快速的辨別商品的好壞一個重要參考。一些研究已提出了根據文字評論來計算星號等級的方法，然而在一些同時有提供文字評論及星號等級的網站中，這種計算出的星號等級與消費者原來提供的有著相當的差異。因此如何讓評論中的情緒傾向正確的分類到所對應的星號等級是很重要的。本論文提出兩種方法，一是透過情緒分析及基因演算法提出一個星等評分的分法，根據中文的餐廳文字評論，計算出一個與消費者所給的較接近的星號等級，另一個則是對某一特定消費者的某一特定評論，我們要判斷這篇評論的星等評論是高於還是低於她(或他)過去所評論過的星號等級之平均。如果高於其平均星號等級，則給予一個正號，反之，則給一個負號，並利用正負標記對餐廳進行評價，提供消費者另一個新的參考指標。
並列摘要
Today's consumers are increasingly inclined to shop on the Internet and comment on the merchandise. Through sentiment analysis, the opinion of the consumer can be abstracted and these comments will affect the willingness of other consumers to buy the same thing. The consumer provides the comments on services or merchandises by giving text, star ratings, or both. Although text comments can give more detailed explanations, consumers can not browse every comment since sometimes there are a huge amount of comments. Hence, the star rating has become an important reference for consumers to quickly identify the quality of products. Some studies have proposed methods for calculating star ratings based on text reviews. However, in some websites that provide both text reviews and star ratings, this calculated star rating is different from what consumers originally provided. Therefore, it is important to correctly classify the emotional tendencies in the comments to the corresponding asterisk level. In this paper we propose two methods. The first is to propose a method for calculating star rating based on sentiment analysis and genetic algorithms. According to Chinese restaurant reviews, calculate a star rating closer to that given by consumers. The other is for a particular comment of a particular consumer, we find whether the consumer's star rating is higher or lower than the average of all her (or his) past star ratings. A positive token is given is this star rating is higher than the average. On the other hand, a negative token is given if it is lower than the average. And use positive and negative tokens to evaluate the restaurant. Finally, we provide consumers with another new reference indicator.