Identifying Sentiment in Web Multi-topic Documents

Most of web documents coverage multiple topic. Identifying sentiment of multi-topic documents is a challenge task. In this paper, we proposed a new method to solve this problem. The method firstly reveals the latent topical facets in documents by Parametric Mixture Model. By focusing on modeling the generation process of a document with multiple topics, we can extract specific properties of documents with multiple topics. PMM models documents with multiple topics by mixing model parameters of each single topic. In order to analyze sentiment of each topic, conditional random fields techniques is used to identify sentiment. Empirical experiments on test datasets show that this approach is effective for extracting subtopics and revealing sentiments of each topic. Moreover, this method is quite general and can be applied to any kinds of text collections.