Blog opinion retrieval based on topic-opinion mixture model

Recently, as blog is becoming a popular medium to express opinions, blog opinion retrieval excites interest in the field of information retrieval. It helps to find and rank blogs by both topic relevance and opinion relevance. This paper presents our topic-opinion mixture model based approach to blog opinion retrieval in the TREC 2009 blog retrieval task. In our approach, we assume each topic has its own opinion relevance model. A topic-opinion mixture model is introduced to update original query model, and can be regarded as a mixture of topic relevance model and opinion relevance model. By pseudo-relevance feedback method, we can estimate these two models from topic relevance feedback documents and opinion relevance feedback documents respectively. Therefore our approach does not need any annotated data to train. In addition, the global representation model is used to represent an entire blog that contains a number of blog posts. Experimental results on TREC blogs08 collection show the effectiveness of our proposed approach.