Feature-Opinion Co-extraction Based Upon Genuine Score Analysis

Abstract

In the recent few years, the Internet has become more user-centric. People prefer to use it for sharing their opinion through different social media. This, in turn, leads to a tremendous increase in the volume of user-generated content. Mining this users’ opinion base is useful from the perspective of buyers as well as sellers. It involves the task of textual analysis. The purpose of this paper is to co-extract the candidate feature-opinion (FO) pairs and find out genuine pairs out of it. In this paper, we focused on rule-based or supervised approach for co-extracting FO pairs. Our main contribution is that we proposed a statistical technique to grab genuine FO pairs. It is a mathematical blend of regularity score (representing an association of an FO pair with the document) and opinion association score (representing an interassociation between the components of FO pair itself). We followed a loosely constrained information extraction unit to design a domain-independent system. To calculate the score values, the candidate components are modeled by bipartite graph and treated by HITS algorithm. For the experiment purpose, we used Opinosis Dataset 1.0. The experimental analysis shows that our mathematical model provides a better result than the existing approach.