Does Reasoning Make Search More Fair? Comparing Fairness in Reasoning and Non-Reasoning Rerankers

Collection of recent IR and recommender-systems papers (arXiv, 12 Mar 2026) covering: (1) fairness comparison showing reasoning-based rerankers do not meaningfully change fairness metrics vs non-reasoning rerankers (Attention-Weighted Rank Fairness stable); (2) causal-attention reformulations for generative recommender systems (AttnLFA, AttnMVP) that cut sequence length and training cost while improving loss/entropy; (3) Differentiable Geometric Indexing (DGI) to make indexing end-to-end differentiable (Gumbel-Softmax, weight sharing) and mitigate popularity/hub bias via scaled cosine on the ℓ