Opinion Driven Decision Support System

Opinions on the web present a wealth of information that can be leveraged in our day to day decision making tasks ranging from which product to purchase to which doctor to consult for a particular ailment. Due to the large volume of opinions available from different sources across web, digesting all the available opinions is a time consuming process which can severely impair user productivity. As a result, these valuable opinions become more of a hindrance than a help in decision making scenarios especially those involving a large number of entities. Most existing work on solving this general problem has been focused on summarizing opinions to help users better digest all the opinions. Unfortunately, in many decision making scenarios, the number of entities in consideration could be quite large. Thus, making decisions by reading summaries alone would still be inefficient as you would need to read summaries of different entities thoroughly. Further, as most of the opinion summarization systems focus on generating summaries that are highly structured, these summaries lack details that can aid decision making. In this thesis, we propose a more efficient way of leveraging opinions, that is to combine the strengths of search technologies with opinion analysis and mining tools to provide a powerful decision making platform. This special platform is called an Opinion-Driven Decision Support System (ODSS) a platform that enables users to find and analyze entities of interest based on opinions of other web users. We study three important problems of the ODSS, encompassing search, analysis and data acquisition. First, in providing a useful search capability, we study the problem of Opinion-Based Entity Ranking where entities are ranked based on a set of user specified opinion preferences. Then, in providing analysis tools to aid decision making, we study Abstractive Summarization of Opinions, where unstructured summaries highlighting key opinions are generated on any arbitrary topic. In order to enable the search and analysis components, opinionated content is imperative. Hence, in the third part of this thesis, we attempt to study the problem of Opinion Acquisition. In the first part of this thesis, we investigate the use of robust retrieval models and extensions of it for the task of Opinion-Based Entity Ranking. Our evaluation, in two different domains, shows that the proposed methods can be directly applied to rank different types of entities for which opinions are available. Our