De-Biasing user conformity bias and item popularity bias in Group recommendation
Junjie Jia, Tianyue Shang, LiFang Li, Si Chen · 2023
The recommendation system is usually faced with the problem of popularity bias, and the collected interactive data often presents quite unbalanced or even long-tail distribution.However, most of the existing studies have considered how to eliminate the negative effects brought by the popularity bias, thus ignoring the two sides of the popularity bias.At the item level, people tend to comment on popular items. So is it all positive? Most research on popularity bias has focused on the conformity behavior and has increased the diversity of personal recommendation lists. At the user level, does everyone want to pursue the preferences of the mainstream group? The reported user responses are skewed toward popular items rather than users’ true interest. We believe that not all popularity bias are harmful. In addition to the consistency effect, the uneven distribution of items can be attributed to the diversity of items’ quality. To address this problem, we propose a new de-biasing mechanism, which uses user conformity bias and item popularity bias to improve the accuracy of recommendation.