The Effect of Document Selection on Query-focused Text Analysis

Collection highlights advances in retrieval, recommendation, and RAG pipelines plus a notable security paper: “AdversarialCoT” demonstrates a practical single-document poisoning attack against retrieval-augmented LLM reasoning. The attack crafts a query-specific adversarial chain-of-thought by iteratively probing the target LLM and refining one poisoned document, and a single such document can substantially degrade reasoning accuracy. Broader themes in the set (document selection effects, opinion-aware retrieval, PDF parsing/chunking, hierarchical indexing, recommendation and cold-start models