Feature sentiment diversification of user generated reviews: The FREuD approach

Abstract

Online discussions, user reviews and comments on the Social Web are valuable sources of information about products, services, or shared contents. The rapidly growing popularity and activity of Web communities raises novel questions of appropriate aggregation and diversification of such social contents. In many cases, users are interested in gaining an extensive overview over pros and cons of a particular track of contributions. We address the problem of social content diversification by combining latent semantic analysis with feature-centric sentiment analysis. Our FREuD approach provides a representative overview of sub-topics and aspects of discussions, characteristic user sentiments under different aspects, and reasons expressed by different opponents. In experiments with real world product reviews we compare FREuD to the typical implementation of ranking reviews by the usefulness rating provided by users as well as a naive sentiment diversification algorithms based on star ratings. To this end we had human users provide a fine-grained gold standard about the coverage of features and sentiments in reviews for several products in three categories. We observed that FREuD clearly outperforms the baseline algorithms in generating a sentiment-diversified set of user reviews for a given product. Copyright © 2013, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved.