| Summary: | 碩士 === 國立中正大學 === 資訊管理學系暨研究所 === 103 === With the rapid development of the Internet, passengers are able to share their travel experience on the global Internet platform. The users are not only receive the information passively, but become the active information disseminator. In such a context, a framework includes “who” says “what” and “how” they say it has been created. Despite the increasing of the user reviews, the differences between quality and helpfulness have caused information overload. The information on the Internet is easy to access but hard to handle and determine. However, a helpful review not only can affect the consumers’ decision, but also influence the enterprises’ sales amount and profit. Therefore, it is an important issue to identify helpful review fast and precisely. This study use Tripadvisor.com as a database for the empirical analysis. The data include all the hotel reviews among five cites in the U.S., namely, New York City, Las Vegas, Chicago, Orlando and Miami. The main purpose of the thesis is to analyze review helpfulness through three features, including review quality, sentiment analysis and reviewer characteristic. Using WEKA data mining software to build predictive models, we evaluate the prediction performance of different classification techniques and different area datasets. We further examine the relative importance of the different feature categories, through select attribute module to understand the most important research variables. The empirical evaluation suggests the reviewer features are shown to have the most impact, RECENCY turned out to be the strongest single predictor. Although the quality and sentiment features are not good as we expected, we can filter out noise variables by feature selection, use a few of the most important variables to construct a good prediction model to help travelers or travel industry to find the most helpful reviews.
|
|---|
