Six tourism themes, diet, entertainment, shopping, view, transportation, and accommodation, are selected for thematic sentiment analysis. 53140 Weibo items published by Chinese tourists in Japan are collected and manually labeled as the case study dataset. Maximum Entropy model and Support Vector Machine are adopted. The training results are both fairly good, where the resulting Maximum Entropy model prevails slightly. It can be concluded that machine learning models are reasonably feasible in tourist thematic sentiment analysis. Moreover, the experiment also shows that the models can be improved by introducing emoticon icons and thematic words as supplements to traditional word features.