Text Chunking Strategies for GraphRAG

Text chunking strategies: Fixed-Size Chunking splits by token count (300 tokens with 100 overlap). Semantic-Aware Chunking splits at paragraph boundaries, improving extraction quality by 30-40%. Recursive Character Splitting tries multiple separators in order. Document-Structure Aware uses headings and lists for coherent chunks. Best practices: Use 300-600 tokens per chunk, include 50-100 token overlap, preserve sentence boundaries, and consider the LLM's context window.