Disentangling interest and conformity for eliminating popularity bias in session-based recommendation

Session-based recommendation (SBR) is to predict the next item, given an anonymous interaction sequence. Recently, many advanced SBR models have shown great recommending performance, but few studies note that they suffer from popularity bias seriously: model tends recommend popular items and fails long-tail items. The only debias works relieve indeed. However, ignore individual’s conformity toward thus decrease performance on Besides, always entangled with real interest, which hinders extracting one’s comprehensive preference. To tackle problem, we propose framework Disentangling InteRest Conformity for eliminating in SBR. In this framework, two groups of item encoders session modeling modules are devised extract interest conformity, respectively, a fusion module designed combine these types Also, discrepancy loss utilized disentangle representation conformity. our can integrate several seamlessly. We conduct extensive experiments three real-world datasets four models. results show outperforms other state-of-the-art methods consistently.