A Counterfactual Approach for Addressing Individual User Unfairness in Collaborative Recommender System

Collection of recent arXiv papers (March 17, 2026) across information retrieval, recommender systems, and LLM applications. Key contributions include: a counterfactual method to mitigate per-user unfairness in collaborative recommenders (introducing synthetic interactions and evaluation on MovieLens/Amazon); an empirical study showing prompt-only LLM query rewriting has domain-dependent effects on dense retrieval and risks reducing lexical alignment; techniques to suppress domain-specific hallucination for regulatory technical standards using QLoRA fine-tuning and GraphRAG with a Neo4j KG (not