INTRODUCTION

With the rapid development of Web2.0 and e-commerce that emphasizes the participation of users, more and more Websites, such as Amazon (http://www.amazon.com) and Epinions (http://www.epinions.com), encourage people to express opinions on products by posting reviews. These reviews are very useful for readers and will possibly influence their purchasing decisions. However, it would cost too much time for readers to read all of the hundreds of reviews of the same product. Thus, automatic review mining and summarization is a very practical concern. The most research on this topic focus on feature-opinion pairs extraction and sentiment orientation decision.

However, a human reader is usually not completely satisfied by a machine generated dull report and may still prefer to read a vivid and complete review article written by a good human writer. This raises the need of selecting the best review from a set of reviews. If only one review from a set was to be read, the most sensible choice would be the most comprehensive review that covers the most information about the other reviews. This is the first goal of our review selection method.

After the most comprehensive review is read, a user may become more interested in one particular feature of the product and would like to read another concise and representative review focusing on that feature only. Therefore it becomes useful to select the best review that focuses on a given feature and represents the other reviewers' opinions on that feature only. This becomes the second goal of our review selection method.

PAPER
Information Shared by Many Objects: information.pdf

DATA SETS
Testing Sets for Comprehensive Review Selection: reviews_for_comprehensive.zip
Annotations for Comprehensive Review Selection: annotations_for_comprehensive.zip
Testing Set for Specialized Review Selection: reviews_for_specialized.txt