Original title: Cracking the Code of Negative Transfer: A Cooperative Game Theoretic Approach for Cross-Domain Sequential Recommendation
Authors: Chung Park, Taesan Kim, Taekyoon Choi, Junui Hong, Yelim Yu, Mincheol Cho, Kyunam Lee, Sungil Ryu, Hyungjun Yoon, Minsung Choi, Jaegul Choo
This article explores Cross-Domain Sequential Recommendation (CDSR), a powerful method using information from multiple domains for accurate and varied recommendations, considering how users interact over time. But there’s a challenge: negative transfer. When knowledge from different domains clashes, it can harm recommendation quality due to varying user tastes. To combat this, the researchers propose a solution. They’ve developed a framework that assesses negative transfer between domains, assigning lower weights to problematic predictions. How? They measure negative transfer using cooperative game theory, evaluating each domain’s impact on model performance. Additionally, they’ve crafted a learning approach that merges coarse-level and fine-level categories, bridging gaps between potentially unrelated fine-level domains by leveraging broader preferences at the category level. Their model outshines previous approaches, demonstrating superior performance across real-world datasets spanning ten diverse domains. Essentially, they’ve found a way to improve recommendations across various domains, minimizing negative effects caused by differences in user preferences.
Original article: https://arxiv.org/abs/2311.13188