Blog Opinion Retrieval with Generation Model and Mixture Model

Blog opinion retrieval aims to find blogs with opinionated information related to a given topic. Its main problem is to compute the opinion score, which balances topic relevance and opinion relevance. To deal with this problem a generative model deduced by a Bayesian approach is pro-posed, and an improved mixture model is proposed to estimate the opinion relevance between a blog and a given topic in our retrieval framework. Moreover, pointwise mutual information is used to expand sentiment words for different topics based on a general sentimental lexicon. The correlation between topic and candidate words is applied in the process of both expanding sentiment words and estimating sentence opinion scores. Experimental results show that the proposed approaches improve upon the state-of-the-art opinion retrieval method on TREC2010 dataset.