Document Chunking
Overview
Document chunking studies how to divide long or complex documents into units that models can retrieve, understand, and process without losing important context. Common approaches include fixed-token or recursive splitting, semantic boundaries, and structure-aware methods that preserve sections, tables, or document hierarchies; newer methods adapt boundaries to likely queries. Evaluations commonly measure retrieval and question-answering quality alongside index size, latency, and token use, and show that parsing and embedding choices interact with chunking. There is no universally best strategy: preserving structure and cross-section evidence can improve results, but may increase cost, while benefits vary by domain, document layout, and task.
Latest papers
54 papers4 papers in the last four weeks, level with the four weeks before. 0.0% of all new papers.