Study Of Aspect Based Sentiment Analysis Based On Graphical Model

With the emergence of various new forms of Internet media and the development of e-commerce platform, the growing user community are generating a large amount of user-generated content in the extensive participation in network activities. Under such a circumstance, accurate emotional analysis and view mining for massively subjective texts on the Internet becomes more and more important, and fine-grained affective analysis, especially aspect-based sentiment analysis techniques, has developed rapidly in recent years. aspect-based sentiment analysis aims to exploit the different sentiment orientation of users in a given commentary,transforming the commentary text of natural language into a structured aspect-sentiment data to support applications that require more granular sentiment analysis, product modeling, user portraits, and personalized recommendations.In this paper, a variety of probabilistic graph models were used to study the aspect-based sentiment analysis in several review datasets. In the study of aspect terms extraction and sentiment word extraction, this paper improves the topic model of LDA in many ways to make it more suitable for the application of sentiment analysis. By utilizing the improved LDA model, a semi-supervised learning model based on CRF is proposed. The aspect term extracted by LDA are labeled as seed words with syntactic features and LDA semantics are introduced. Compared with the traditional non-supervised learning methods, the precision of aspect terms extraction have been greatly improved. By introducing the user-product distribution and product-aspect distribution according to the production process of the actual comments, a user-product probabilistic model is proposed, and the rating of this comment is introduced as a observed variable. Combined with a community-based discovery model based on random blocks, this model can train individual models for different communities of users and compare models using the results of product recommendation algorithms.