Modeling on evaluation object extraction in e-commerce corpus based on semantic feature

摘要

As a newly shopping tool, electronic commerce has been drawing more and more attention of researchers. According to the characteristics of comments diversity, it is necessary to extract evaluation object which is an important component of sentiment information. This paper explores Conditional Random Field (CRF) to do evaluation objects extraction. After observing generally used features in sentiment extraction, this paper conclude all the features into four categories, i.e. word Segmentation, Part-of-speech Tagging (POS), Dependency Parsing, Semantic Dependency Parsing. What's more, focusing on the introduction of new feature semantic dependency is a very vital item in our research. In the experiment, we examine the various features and combinations in the extraction task performance, and make a detailed comparative study. The experimental results confirm that adding the feature of semantic dependency has better performance in terms of the evaluation object extraction.