Adaptive Retrieval Strategies for Biomedical Question Answering

This feed contains multiple 2026 IR/ML papers focused on retrieval-augmented QA, recommender systems, LLM internal representations, task decomposition, and fairness in algorithmic matching. Highlights include an adaptive retrieval framework for biomedical QA that tailors retrieval and aggregation strategies by question type (evaluated on BioASQ); a study showing mixed effects of giving users control over news recommender filter bubbles; layer-wise linear probes demonstrating that LLMs encode query-document relevance in middle-to-late transformer layers (TREC DL20, MIRACL); RecGPT-V3, a statefu