Web Retrieval-Aware Chunking (W-RAC) for Efficient and Cost-Effective Retrieval-Augmented Generation Systems

Collection of recent arXiv CS/IR papers focused on retrieval-augmented generation (RAG), retrieval for agentic systems, document conversion for RAG, multimodal graph RAG, recommendation (including large-scale advertising datasets), unlearning for LLM-based recommenders, bias mitigation in generative recommendation, and evaluation/comparison of embedding vs generative classifiers. Key technical contributions include Web Retrieval-Aware Chunking (W-RAC) to reduce LLM token costs and hallucination risk for web content ingestion; LRAT for mining supervision from agent trajectories; MG^2-RAG for a