Module semantic_coherence 

Semantic coherence scoring for BAR-RAG Semantic Coherence Scoring for Boundary-Aware Chunking

This module implements semantic coherence analysis using sentence embeddings to optimize chunk boundaries for maximum semantic unity.

Key capabilities:

• Cosine similarity calculation between sentence embeddings
• Intra-chunk coherence scoring
• Optimal split-point detection via binary search
• Adaptive threshold based on embedding distances

References

• BAR-RAG Paper: “Boundary-Aware Retrieval-Augmented Generation”
• Target: +40% semantic coherence improvement

Structs

CoherenceConfig

Configuration for semantic coherence scoring

OptimalSplit

Result of split-point optimization

ScoredChunk

Represents a candidate chunk with coherence score

SemanticCoherenceScorer

Semantic coherence scorer using sentence embeddings