Web Forum Sentiment Analysis Based on Topics

As Web forum has become an enormous collection of highly valuable opinions and commentaries, more and more researchers express strong interests on it. However, most of them pay attention to the forum reviews rather than the posts themselves. In this paper we focus on recognizing the diversified opinions of different threads on the same topic from Chinese Web forums. First, we congregate the Web forum threads on the same topic into a cluster. In order to further distinguish different opinions for a topic, we propose a sentiment classification algorithm based on a probability word-list, which is constructed by us using a propagation method on the word graph with a seed set. Experimental results on a real data set show that our algorithm performs well with both high precision and efficiency, much better than traditional methods such as SVM and naive Bayes.