Detecting reviewer bias through web-based association mining
Jessica Staddon, Richard Chow · 2008
Online retailers and content distributors benefit from an active community that shares credible reviews and recommendations. Today, the most popular approach to encouraging credibility in these communities is self-regulation; community members rate reviews according to their accuracy and usefulness, thus helping to weed out reviews that are inaccurate. This self-regulation, while powerful, is limited by its insularity. Commu-nity members generally base their assessments on a re-viewer’s comments and actions only within the commu-nity. This ignores relationships the reviewer has outside the community that may be quite relevant to evaluat-ing the reviewer’s comments; for example, a relationship between an author and reviewer. We present a simple method for mining the Web to detect many such as-sociations. Our method, together with self-regulation, provides for more comprehensive detection of bias in re-views by alerting the user to the potential for an undis-closed relationship between a reviewer and author. We provide preliminary results using book reviews in Ama-zon.com demonstrating that our approach is a high-precision method for detecting strong relationships be-tween reviewers and authors that may contribute to re-viewer bias.