Topic-Sentiment Mining from Multiple Text Collections

摘要

Topic-sentiment mining is a challenging task for many applications. This paper presents a topic-sentiment joint model in order to mine topics and their sentimental polarities from multiple text collections. Text collections are represented with a mixture of components and modeled via the hierarchical Dirichlet process which can determine the number of components automatically. Each component consists of topic words and its sentiments. The model can mine topics with different proportions and different sentimental polarities as well as one positive and one negative topic for each collection. Experiments on two text collections from Chinese news media and microblog show that our model can find meaningful topics and their different sentimental polarities. Experiments on Multi-Domain Sentiment Dataset show that our model is better than the JST-alike models on parameter settings for topic-sentiment mining.