摘要：Sentiment analysis or opinion mining in online product reviews is a useful method that can automatically detect subjective information regarding the entity such as opinions, attitudes, and feelings expressed by consumers. Online product reviews always include objective and subjective sentences; identification of subjective sentences in the given content is a very important and foundational task in the research of opinion mining. In this paper, we focus on the problem of identification of sentence-level subjective sentences, propose a weakly supervised model mixed topics based on LDA for identification of the subjective sentences, considering the impact of multiple topic factors on the identification of subjective sentences. The approach exploits semi-supervised learning method, and extended the existing basic LDA topic model for the identification of subjectivity in text. This work iterates the model prior probability by using a small domain-independent lexicon. Finally, the proposed model is applied to a online review corpus and the experimental shows that the proposed model can effectively improve the recognition effect.
