SentenceChunkingStrategy
Models & Types · AiDotNet.RetrievalAugmentedGeneration.ChunkingStrategies
Splits text into chunks based on sentence boundaries to preserve semantic coherence.
Inheritance: ChunkingStrategyBase → SentenceChunkingStrategy
For Beginners
This keeps complete sentences together in each chunk.
Think of it like organizing a book:
- Bad way: Cut every 500 characters, even mid-sentence
“The cat sat on the m|at. The dog ran ar|ound the yard.”
- Good way: Keep sentences whole
Chunk 1: “The cat sat on the mat. The dog ran around the yard.” Chunk 2: “The bird flew over the fence. The fish swam in the pond.”
Why this matters:
- Retrieval works better when searching complete thoughts
- Generators get more coherent context
- No weird sentence fragments that confuse the model
Parameters:
- targetChunkSize: Aim for this many characters per chunk
- maxChunkSize: Never exceed this size (may break sentences if needed)
- overlapSentences: Number of sentences to repeat between chunks for context
Example with targetChunkSize=100, overlapSentences=1: “First sentence. Second sentence. Third sentence. Fourth sentence.”
Chunk 1: “First sentence. Second sentence. Third sentence.” Chunk 2: “Third sentence. Fourth sentence.” (overlap: “Third sentence”)
How It Works
This chunking strategy splits text at sentence boundaries (periods, question marks, exclamation points) and combines sentences until reaching the target chunk size. This approach preserves complete thoughts and improves retrieval quality compared to arbitrary character-based splitting.
Constructors
| Constructor | Summary |
|---|---|
SentenceChunkingStrategy(int targetChunkSize, int maxChunkSize, int overlapSentences) |
Initializes a new instance of the SentenceChunkingStrategy class. |