Debiasing Message Passing to Mitigate Popularity Bias in GNN-based Collaborative Filtering

Collection of arXiv papers (2026-05-13) focused on modern information retrieval and recommendation research. Key contributions include DPAA: an embedding-aware, layer-wise debiasing method for popularity amplification in GNN-based collaborative filtering; MIRA: a large-scale, LLM-assisted multi-category retrieval benchmark built from real user queries; Tenure: a local-first typed belief store for cross-session LLM memory with scope isolation and versioned epistemic status; a large-scale analysis and taxonomy showing much web geospatial search lies outside traditional GIS; FedMM: a privacy-pres