Sentiment analysis of online product reviews with Semi-supervised topic sentiment mixture model

Analysis the positive and negative sentiments about each topic of the product are very useful to the customers and manufacturers. In this paper we propose a new topic sentiment mixture model which we call Semi-supervised Co-LDA model to obtain the positive and negative opinions from the reviews about each product. The Semi-supervised Co-LDA can model the topic and sentiment of the product reviews simultaneously. The Semi-supervised Co-LDA model we proposed is a semi-supervised model, which utilizes the well-written expert reviews as labeled data. The Co-LDA model has an additional advantage that can integrate expert opinions and ordinary opinions. Empirical experiments on the online reviews datasets from CNET show that this approach is effective for topic sentiment analysis of the product. The Co-LDA model is quite general, which can be applied to many fields such as modeling opinions in weblogs, user behavior prediction.