Do LLM Recommenders Know When They're Hallucinating? Auditing Confidence Calibration in Catalog Faithfulness

This arXiv feed contains new research on recommender systems and information retrieval, including LLM hallucination and confidence calibration, social-graph personalization, robustness to missing modalities, sign-aware and spectral collaborative filtering, LLM-backed ranking at Netflix, curvature-aware graph propagation, persona sensitivity in LLM evaluation, reliability reranking, and efficient reasoning via model merging. The material is academic and does not describe an actively exploitable security vulnerability.