부산대학교 도서관

E-commerce is on a boom and consumer reviews have become an essential aspect of this growing business. Various models have been proposed that use user feedback reviews to compute seller trust profile. The feedback reviews consist of ratings as well as comments given by the user. Models that compute seller trust based on user feedback ratings overlook the important subjective details in the text and give inaccurate trust scores based on ratings which are objective in nature. We have proposed a novel model for computing seller trust profile based on fine grained analysis of feedback comments. In our approach, we have proposed a unique methodology to compare and verify our result with two e-commerce websites. A few common seller on Amazon and Flipkart have been identified and trust scores have been computed for them. The computed trust scores for the sellers on Amazon are considerably lesser than those displayed on the website while those for Flipkart are nearly equal to those mentioned on its website. The calculated relative error is considerably higher on Amazon as compared to Flipkart. The experiment is unique of its kind in the field of natural language processing and sentiment analysis, and achieves the stated objective.