RWKV4Rec: RWKV-Based Personalized Sequential Recommendation Model

Abstract

Sequential recommender systems face high resource consumption and inefficient processing as user behavior sequences grow longer. RWKV4Rec adapts the RWKV architecture to sequential recommendation with linear-complexity recurrent processing, an Item-RWKV block, and a Low-Rank Time Mix module based on LoRA. The model captures long-term user preferences while reducing computational cost, and improves recommendation performance across benchmark datasets.