A Proposal of a Method to Automatically Estimate Evaluations of Various Topics of Travelers' Reviews

Generally, evaluations in customer reviews are shown with text and score. In addition, there are 2 kinds of scored evaluations, which are a comprehensive evaluation and topic evaluations. However, topic evaluations are not often given by reviewers. Therefore, it is required to estimate the evaluations. In this paper, we researched a performance of estimation by SVM. Furthermore, we compared results of estimation by SVM and scanning with dependency relation tree. The result shew that precision of estimation by each method is low. On the other hand, the result shew that precision in positive score is over about 70% in each topic.