Analyzing Reputation by Mining Feedback Comments

Generally online (Electronic commerce or E-commerce) applications use reputation reporting system for trust evaluation where they gather overall feedback ratings from the sellers to compute the reputation score. A well-known issue with the reputation conduct system is " all good reputation " problem where over 99% of feedback ratings are positive leading to high reputation scores. This issue is hard on buyers to select accurate sellers. By analyzing buyer's opinions on free text feedback comments, we propose an approach called the Reputation Analyzer. The main idea behind reputation analyzer is an algorithm lexical-LDA (Latent Dirichlet Allocation) topic modeling technique proposed for mining the online feedback comments by grouping aspect expressions into dimensions and compute dimension ratings. Extensive experiments on eBay and Amazon data show that the reputation analyzer can significantly solve the " all good reputation " problem and rank sellers effectively.