Sentiment Analysis of Restaurant Reviews on Review Web Sites

Abstract

         Review websites have widely used for searching restaurant information that enables user decision supporting. Unfortunately, the review comments are text messages. Thus, users have to read all comments of the interested restaurants and then summarize some issues. This process is time consuming. Restaurant managers also found that it is difficult to extract the customers’ critics from these comments. This paper proposes a feature-based sentiment analysis system for automatic summarizing customers’ comments. Natural Language Processing (NLP) techniques, including tokenization, analyzing part of speech and sentence pattern analysis, were used for determining the polarity of sentences. Finally a polarity score of each feature was computed and displayed in graphic visualization. The user evaluation was conducted to find user satisfaction on correctness of the sentiment analysis. The result shows that users provide high satisfaction on the analysis for the food and service features, while others reach the fair level of satisfaction.