sentence-transformers-semantic-chunker

This model borrows from Greg Kamradt’s work here: https://github.com/FullStackRetrieval-com/RetrievalTutorials/blob/main/5_Levels_Of_Text_Splitting.ipynb. The idea is to segment text into semantically coherent chunks. The primary goal of this work is to use sentence transformers embeddings to represent the meaning of sentences and detect shifts in meaning to identify potential breakpoints between chunks.

Model Description

This model aims to segment a text into semantically coherent chunks. It uses sentence-transformers embeddings to represent the meaning of sentences and detect shifts in the meaning to identify potential breakpoints between chunks. There are two primary changes in function from Greg Kamradt excellent original work: 1) Introduce the use of sentence-transformer embeddings rather than OpenAI to provide an entirely open source implementation of semantic chunking, and 2) add functionality to merge smaller chunks with their most semantically similar neighbors to better normalize chunk size. The goal is to use semantic Understanding to enable a model to consider the meaning of text segments rather than purely relying on punctuation or syntax, and to provide flexiblity so that the breakpoint_percentile_threshold and min_chunk_size can be adjusted to influence the granularity of the chunks.

General Outline:

Preprocessing

• Loading Text: Reads the text from the specified path.
• Sentence Tokenization: Splits the text into a list of individual sentences using nltk's sentence tokenizer.

Semantic Embeddings

• Model Loading: Loads a pre-trained Sentence Transformer model ( in this case: 'sentence-transformers/all-mpnet-base-v1').
• Embedding Generation: Converts each sentence into an embedding to represent its meaning.

Sentence Combination:

• Combines each sentence with its neighbors to form slightly larger units, helping the model understand the context in which changes of topic are likely to occur.

Breakpoint Identification

• Cosine Distance: Calculates cosine distances between embeddings of the combined sentences. These distances represent the degree of semantic dissimilarity.

• Percentile-Based Threshold: Determines a threshold based on a percentile of the distances (e.g., 95th percentile), where higher values indicate more significant semantic shifts.

• Locating Breaks: Identifies the indices of distances above the threshold, which mark potential breakpoints between chunks. Chunk Creation:

• Splitting at Breakpoints: Divides the original sentences into chunks based on the identified breakpoints.

Chunk Merging:

• Minimum Chunk Size: Defines a minimum number of sentences to consider a chunk valid.
• Similarity-Based Merging: Merges smaller chunks with their most semantically similar neighbor based on cosine similarity between chunk embeddings.

Output:

• The model ultimately produces a list of text chunks (chunks), each representing a somewhat self-contained, semantically cohesive segment of the original text.

Usage

Using this chunker is easy when you have sentence-transformers installed:

pip install -U sentence-transformers

Then you can implement like this:

"""
Text Chunking Utility

This module provides functionality to intelligently chunk text documents into semantically coherent sections
using sentence embeddings and cosine similarity. It's particularly useful for processing large documents
while maintaining contextual relationships between sentences.

Requirements:
    - nltk
    - sentence-transformers
    - scikit-learn
    - numpy
    - matplotlib
"""

import nltk
from nltk.tokenize import sent_tokenize
from sentence_transformers import SentenceTransformer
from sklearn.metrics.pairwise import cosine_similarity
import numpy as np
import matplotlib.pyplot as plt


class TextChunker:
    def __init__(self, model_name='sentence-transformers/all-mpnet-base-v1'):
        """Initialize the TextChunker with a specified sentence transformer model."""
        self.model = SentenceTransformer(model_name)

    def process_file(self, file_path, context_window=1, percentile_threshold=95, min_chunk_size=3):
        """
        Process a text file and split it into semantically meaningful chunks.
        
        Args:
            file_path: Path to the text file
            context_window: Number of sentences to consider on either side for context
            percentile_threshold: Percentile threshold for identifying breakpoints
            min_chunk_size: Minimum number of sentences in a chunk
            
        Returns:
            list: Semantically coherent text chunks
        """
        # Process the text file
        sentences = self._load_text(file_path)
        contextualized = self._add_context(sentences, context_window)
        embeddings = self.model.encode(contextualized)
        
        # Create and refine chunks
        distances = self._calculate_distances(embeddings)
        breakpoints = self._identify_breakpoints(distances, percentile_threshold)
        initial_chunks = self._create_chunks(sentences, breakpoints)
        
        # Merge small chunks for better coherence
        chunk_embeddings = self.model.encode(initial_chunks)
        final_chunks = self._merge_small_chunks(initial_chunks, chunk_embeddings, min_chunk_size)
        
        return final_chunks

    def _load_text(self, file_path):
        """Load and tokenize text from a file."""
        with open(file_path, 'r', encoding='utf-8') as file:
            text = file.read()
        return sent_tokenize(text)

    def _add_context(self, sentences, window_size):
        """Combine sentences with their neighbors for better context."""
        contextualized = []
        for i in range(len(sentences)):
            start = max(0, i - window_size)
            end = min(len(sentences), i + window_size + 1)
            context = ' '.join(sentences[start:end])
            contextualized.append(context)
        return contextualized

    def _calculate_distances(self, embeddings):
        """Calculate cosine distances between consecutive embeddings."""
        distances = []
        for i in range(len(embeddings) - 1):
            similarity = cosine_similarity([embeddings[i]], [embeddings[i + 1]])[0][0]
            distance = 1 - similarity
            distances.append(distance)
        return distances

    def _identify_breakpoints(self, distances, threshold_percentile):
        """Find natural breaking points in the text based on semantic distances."""
        threshold = np.percentile(distances, threshold_percentile)
        return [i for i, dist in enumerate(distances) if dist > threshold]

    def _create_chunks(self, sentences, breakpoints):
        """Create initial text chunks based on identified breakpoints."""
        chunks = []
        start_idx = 0
        
        for breakpoint in breakpoints:
            chunk = ' '.join(sentences[start_idx:breakpoint + 1])
            chunks.append(chunk)
            start_idx = breakpoint + 1
            
        # Add the final chunk
        final_chunk = ' '.join(sentences[start_idx:])
        chunks.append(final_chunk)
        
        return chunks

    def _merge_small_chunks(self, chunks, embeddings, min_size):
        """Merge small chunks with their most similar neighbor."""
        final_chunks = [chunks[0]]
        merged_embeddings = [embeddings[0]]
        
        for i in range(1, len(chunks) - 1):
            current_chunk_size = len(chunks[i].split('. '))
            
            if current_chunk_size < min_size:
                # Calculate similarities
                prev_similarity = cosine_similarity([embeddings[i]], [merged_embeddings[-1]])[0][0]
                next_similarity = cosine_similarity([embeddings[i]], [embeddings[i + 1]])[0][0]
                
                if prev_similarity > next_similarity:
                    # Merge with previous chunk
                    final_chunks[-1] = f"{final_chunks[-1]} {chunks[i]}"
                    merged_embeddings[-1] = (merged_embeddings[-1] + embeddings[i]) / 2
                else:
                    # Merge with next chunk
                    chunks[i + 1] = f"{chunks[i]} {chunks[i + 1]}"
                    embeddings[i + 1] = (embeddings[i] + embeddings[i + 1]) / 2
            else:
                final_chunks.append(chunks[i])
                merged_embeddings.append(embeddings[i])
        
        final_chunks.append(chunks[-1])
        return final_chunks


def main():
    """Example usage of the TextChunker class."""
    # Initialize the chunker
    chunker = TextChunker()
    
    # Process a text file
    file_path = "path/to/your/document.txt"
    chunks = chunker.process_file(
        file_path,
        context_window=1,
        percentile_threshold=95,
        min_chunk_size=3
    )
    
    # Print results
    print(f"Successfully split text into {len(chunks)} chunks")
    print("\nFirst chunk preview:")
    print(f"{chunks[0][:200]}...")


if __name__ == "__main__":
    main()

Evaluation Results

Testing on various buffer sizes and breakpoints using the King James Version of the book of Romans (Available here: https://quod.lib.umich.edu/cgi/k/kjv/kjv-idx?type=DIV1&byte=5015363).

Intra-chunk similarity (how similar the sentences in a given chunk are to each other. Higher = More Semantically Similar):

Inter-chunk similarity (how similar the respective chunks are to each other. Lower = Less Semantically Similar):

Citing and Authors

If you find this model helpful, please enjoy and give all credit to Greg Kamradt for the idea.