tsensei / Semantic-Chunking-Typescript

Typescript implementation of algorithm to split the text based on semantic similarity by Greg Kamradt

Semantic Chunker 💫

This TypeScript project implements an algorithm to split large text corpora into semantically cohesive chunks using embeddings.

Taken from Greg Kamradt’s wonderful notebook: 5_Levels_Of_Text_Splitting

Key Features:

• Intelligent Sentence Grouping: Combines sentences contextually for more meaningful analysis.
• OpenAI Sentence Embeddings: Leverages OpenAI's embedding models to understand text semantics.
• Cosine Similarity Analysis: Measures the semantic 'distance' between sentence groups to pinpoint shifts in topics.
• Flexible Thresholding: Adjust sensitivity to define what constitutes a significant semantic shift.

Getting Started

Clone the repository:

git clone https://github.com/tsensei/Semantic-Chunking-Typescript.git

Install dependencies:

pnpm install

Set up your OpenAI API key:

• Create a .env file by copying .env.example
• Add your OpenAI API key in the .env file

Run the chunker:

tsc
node build/app.js

Customization :

• Experiment with the bufferSize in the structureSentences function to control the contextual window for embeddings.
• Adjust the percentileThreshold in calculateCosineDistancesAndSignificantShifts to fine-tune the sensitivity of chunk boundaries.