Product Feature Summarization by Incorporating Domain Information

Feature summarization is an important problem in opinion mining of product review. Current methods mostly cluster feature expressions with unsupervised learning methods based on lexical similarity or context information similarity. Although several semi-supervised methods have been proposed to addressing the problem, their labeled set needs manual definition and requires a professional knowledge to categorize features into same aspect or different aspects. In this paper, we proposed a semi-supervised method that incorporates domain information in web sites to generate labeled set and constructs a novel context information modeling process with EM method to solve the problem. Experimental results show that the proposed method achieves better performance than existing approaches.