Reproduce “Reinforcing User Interest Evolution in Multi-Scenario Learning for recommender systems” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Zhijian Feng, Wenhao Zheng, Xuanji Xiao
A framework called RUIE is presented, which uses reinforcement learning to model the dynamic evolution of user interests across various scenarios in recommender systems. This approach addresses limitations of existing models by incorporating Double Q-learning and a Q-value-gated triplet loss, leading to a 229% improvement in NDCG@20 over the best multi-scenario baseline on the KuaiSAR dataset.
Key insights
- Reinforcement learning, specifically Double Q-learning, can effectively model the dynamic evolution of user interests across diverse interaction scenarios in recommender systems.
- A Q-value-gated contrastive loss can dynamically adjust learning weights to mitigate negative knowledge transfer between scenarios, ensuring beneficial cross-scenario learning.
- Multi-Head Attention is crucial for capturing the migration and evolution of user interests by learning associations between items and their interaction contexts.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.