Context sensitive contrastive feature based opinion summarization of online reviews

Abstract

The computerized era has enabled the general public to convey their views about product preferences on review websites, blogs, and forums. These views are a significant source for prospective customers to make a purchase decision and organizations to mold their product. However, choosing appropriate information from this enormous quantity of views needs the customer to spend considerable time reading reviews and digesting contradictory views published by various reviewers. These emerging needs have inspired a new area of research on mining customer reviews, opinion mining. Opinion mining deals with extracting knowledge from the opinionated text about a particular topic using methods of machine learning, Natural language processing, and Linguistics. Contrastive summarization is an emerging task in opinion mining that aggregates and presents the opinions in a form that would help the users to make better decisions. Conventional summarization systems are confronted by a few limitations in different stages of summary generation. It includes lack of techniques to address semantic redundancy problems in extracting feature-opinion phrases. Non-availability of a method for extracting and incorporating contexts in determining the implicit opinion present in a sentence. Also, an end-to-end solution for automatic contrastive summarization is not investigated. This work aims to overcome the above-mentioned limitations of summarization using recent developments in natural language processing methods. A new framework for contrastive summarization using a model-based approach is presented in this thesis newline