Opinion Searching in Multi-Product Reviews

It is becoming common that people browse Web for product reviews before purchasing. However, to retrieve opinions relevant to customer desire still remains challenging. In this paper, we studied the problem of opinion searching, whose aim is to search the opinions about specific feature of specific product and locate them in multi-product reviews. Our solution includes two steps: opinion indexing and opinion retrieving. Opinion indexing is to identify opinion fragments and generate opinion tuples (product,feature and sentiment). Opinion retrieving is to look up the opinion tuples matching users' retrieving interests, and help users to locate the corresponding opinion fragments in documents. Fundamentally, opinion indexing should be able to identify the feature-product dependencies (i.e., a feature mentioned in somewhere of reviewing text is semantically associated with which product). We explore to resolve the problem with machine-learning techniques.