Sentiment Analysis Of Travel Destination Reviews In Chinese

With the rapid development of the Internet, the web has become a convenient platform for information exchange. More and more people begin to publish information or their opinions on the Internet. And travel virtual community is increasingly popular for tourists, because the contents of reviews contributed by common tourists are deemed to be reliable, which are always taken into consideration while people choose travel destinations. Actually, mining the information in the abundant online travel-related reviews can help to understand the opinions or evaluations from customers effectively and then to improve products or services, so that it becomes one of the key to the success of tourism e-commerce.Sentiment classification aims to mining the customers'reviews on the web comprehensively and efficiently by classifying the reviews into positive or negative opinions. At present, there have been many studies on sentiment analysis for English traveler reviews, and obtained some results. As we know, China is the largest country in terms of the number of Internet users, and Chinese information has become a very important part on the Internet. But in this field, special challenges are associated with the mining of Chinese reviews. To solve the problem, this study conducts an exploring research on sentiment analysis to Chinese traveler reviews.The experiment data of Chinese reviews for travel destination are automatically downloaded from www.ctrip.com by programming. This study conducted an exploring research on sentiment analysis model by four algorithms, including the Point-wise Mutual Information method and three popular supervised machine learning algorithms:Support Vector Machine, Naive Bayes and N-gram method. Then, we compared the performances of the four classifiers applied to online Chinese reviews generated by travelers. Empirical results indicate that, when the size of the training set is different, the three classifiers based on supervised machine learning algorithms perform differently. As a whole, SVM algorithm can gain a better performance, which accuracy is close 90%. Comparing to classifiers based on machine learning algorithms, the classifier based on PMI method doesn't perform better, which accuracy is just a little more than 80%. But the PMI semantic-oriented method is likely more applied in practice, because it requires less laborious hours on the project.Finally, based on the certain sentiment phrases extracted in the PMI experiment, the paper presents and analyzes the existing problems of travel destinations from some aspects, including tour, entertainment, food, accommodation, shopping and traffic, and then gives some suggestions. Additionally, we put forward the potential value of applying sentiment analysis technologies to tourism electronic commerce system.