Query-Adaptive Semantic Chunking for Retrieval-Augmented Generation: A Dynamic Strategy with Contextual Window Expansion

In three linesQASC (Query-Adaptive Semantic Chunking) improves document segmentation for RAG by integrating user queries at chunking stage. Using cosine similarity scoring, contextual window expansion, and chunk-level aggregation, QASC achieves F1=0.85, a 18-27% relative improvement over fixed chunking and 8-12% over semantic/agentic methods on 100 technical documents and 200 queries.

Summary generated by Claude — human-verified