Lexicon construction: A topic model approach

Sentiment Analysis has been an interesting task of web content mining these years due to rapid growth of user generating content. As the annotated data are expensive to get, the unsupervised approaches are preferred. Usually, a lexicon is required when apply the unsupervised approaches. In the paper, based on Latent Dirichlet Allocation (LDA), we propose a model to construct a lexicon for sentiment analysis task, which is domain independent. Through experiments, we compare our generated lexicon with some widely used lexicons and with trivial lexicon construction algorithm. The experiments show our approach is competitive and flexible.