Relevance Matters: A Multi-Task and Multi-Stage Large Language Model Approach for E-commerce Query Rewriting

Collection of recent arXiv papers (Mar 4, 2026) in information retrieval and recommender systems presenting methods to improve relevance, scalability, and deployability. Key contributions include a multi-task, multi-stage LLM-based query-rewriting framework with relevance tagging and GRPO fine-tuning deployed on JD.com; SVD-Attention and SOLAR for low-rank, long-sequence recommendation at Kuaishou; FlashEvaluator for sublinear, cross-sequence evaluation; AlphaFree removing user/ID/GNN dependencies via language representations; S2CDR diffusion-inspired cross-domain recommendation for cold-start