ClusterKW

A Node.js package for clustering keywords using OpenAI embeddings. This package can be used to cluster Google Ads keywords, AI chat topics, project tasks, or any other text-based items into semantically similar groups.

Features

• Generate embeddings for keywords using OpenAI API
• Calculate distances between keywords
• Create clusters of semantically similar keywords
• Automatically generate cluster names and descriptions using OpenAI API
• Multiple clustering algorithms:
  • Simple: Basic distance-based clustering
  • K-means: Centroid-based clustering with configurable number of clusters
  • Hierarchical: Agglomerative clustering with different linkage methods
  • Direct: LLM-based clustering without embeddings (sends keywords directly to the model)

Installation

npm install clusterkw

Usage

import { KeywordClusterer } from 'clusterkw';

// Initialize with your OpenAI API key
const clusterer = new KeywordClusterer({
  apiKey: 'your-openai-api-key',
  // Optional configuration
  embeddingModel: 'text-embedding-3-small',
  completionModel: 'gpt-4o-mini-2024-07-18',
  minClusterSize: 3,
  distanceThreshold: 0.2,
  algorithm: 'direct',
  context: 'AI chat topics'
});

// Example keywords
const keywords = [
  'machine learning algorithms',
  'deep learning frameworks',
  'neural networks',
  'data visualization tools',
  'business intelligence software',
  'data analytics platforms',
  'artificial intelligence applications',
  'natural language processing',
  'computer vision systems',
  'reinforcement learning'
];

// Cluster the keywords
const clusters = await clusterer.clusterKeywords(keywords);

// Output the clusters with their names and descriptions
console.log(clusters);

API Reference

KeywordClusterer

The main class for clustering keywords.

Constructor Options

• apiKey (string, required): Your OpenAI API key
• embeddingModel (string, optional): The OpenAI model to use for embeddings. Default: 'text-embedding-3-small'
• completionModel (string, optional): The OpenAI model to use for generating cluster names and descriptions. Default: 'gpt-3.5-turbo'
• minClusterSize (number, optional): Minimum number of items to form a cluster. Default: 2
• distanceThreshold (number, optional): Maximum cosine distance for items to be considered similar. Default: 0.3
• algorithm (string, optional): Clustering algorithm to use ('simple', 'kmeans', or 'hierarchical'). Default: 'simple'
• k (number, optional): Number of clusters for k-means algorithm. Default: auto-calculated based on dataset size
• maxIterations (number, optional): Maximum iterations for k-means algorithm. Default: 100
• linkage (string, optional): Linkage method for hierarchical clustering ('single', 'complete', or 'average'). Default: 'average'

Methods

• clusterKeywords(keywords: string[]): Promise<Cluster[]>: Clusters the provided keywords and returns the clusters with names and descriptions.
• getEmbeddings(texts: string[]): Promise<number[][]>: Gets embeddings for the provided texts.
• calculateDistances(embeddings: number[][]): number[][]: Calculates the cosine distances between all embeddings.
• generateClusters(keywords: string[], distances: number[][], embeddings?: number[][]): Cluster[]: Generates clusters based on the selected algorithm.
• nameAndDescribeClusters(clusters: Cluster[]): Promise<Cluster[]>: Generates names and descriptions for the clusters.

Command Line Interface (CLI)

The package includes a CLI tool for clustering keywords directly from the command line:

# Install globally
npm install -g clusterkw

# Basic usage with a file
clusterkw --file keywords.txt

# Using a CSV file with a specific column
clusterkw --file keywords.csv --column keyword

# Specify output file
clusterkw --file keywords.txt --output clusters.json

# Customize clustering parameters
clusterkw --file keywords.txt --min-cluster-size 3 --distance 0.25

# Use different clustering algorithms
clusterkw --file keywords.txt --algorithm kmeans --clusters 5
clusterkw --file keywords.txt --algorithm hierarchical --linkage complete
clusterkw --file keywords.txt --algorithm direct --gpt-model gpt-4o-mini-2024-07-18

# Use context to guide clustering
clusterkw --file keywords.txt --context "AI chat topics"
clusterkw --file keywords.txt --algorithm direct --context "Google Ads keywords"

# Provide API key directly (alternatively, set OPENAI_API_KEY or OPENAI_KEY environment variable)
clusterkw --file keywords.txt --api-key your-openai-api-key

CLI Options

| Option | Alias | Description | Default | |--------|-------|-------------|---------| | --file | -f | Path to file containing keywords (supports txt, csv, json) | (required) | | --column | -c | Column name containing keywords (for CSV files) | keyword | | --output | -o | Output file path (supports json, csv) | (none) | | --api-key | | OpenAI API key (overrides environment variables) | (from env) | | --min-cluster-size | -m | Minimum cluster size | 2 | | --distance | -d | Maximum distance threshold | 0.3 | | --embedding-model | -e | OpenAI embedding model | text-embedding-3-small | | --gpt-model | -g | OpenAI completion model | gpt-4o-mini-2024-07-18 | | --algorithm | -a | Clustering algorithm (simple, kmeans, hierarchical, direct) | simple | | --clusters | -k | Number of clusters for k-means algorithm | 5 | | --max-iterations | | Maximum iterations for k-means algorithm | 100 | | --linkage | | Linkage method for hierarchical clustering | average | | --context | | Context description to guide clustering | (none) | | --delimiter | | CSV delimiter | , | | --no-header | | CSV file has no header row | (false) |

Version Management

This package follows Semantic Versioning (SemVer):

• MAJOR: Incompatible API changes
• MINOR: Add functionality (backward-compatible)
• PATCH: Bug fixes (backward-compatible)

Updating the Version

You can update the version using npm scripts:

# Increment patch version (1.0.0 -> 1.0.1)
npm run version:patch

# Increment minor version (1.0.0 -> 1.1.0)
npm run version:minor

# Increment major version (1.0.0 -> 2.0.0)
npm run version:major

Automatic Publishing

This package uses GitHub Actions to automatically publish to npm when:

1. A new version is pushed to the main branch (package.json changes)
2. A new GitHub Release is created
3. The publish workflow is manually triggered

License

MIT