Bias disparity in graph-based collaborative filtering recommenders
Georgios Boltsis, Evaggelia Pitoura · Proceedings of the 37th ACM/SIGAPP Symposium on Applied Computing · 2022
In collaborative filtering (CF) recommendations, user preferences may be reinforced resulting in the under-representation of specific categories of items in the recommendation lists received by users belonging to specific groups. We call this phenomenon bias disparity. In this paper, we focus on CF recommenders that exploit the bipartite user-item interaction graph and present algorithms that modify the random walks on this graph to address bias disparity. We evaluate experimentally both the conditions under which bias disparity appears and the effectiveness of our algorithms.