📌 Intent Classifier - AI-powered Intent Recognition

🚀 Overview

Intent Classifier is a lightweight and efficient library for classifying user queries into predefined intents. It supports both TF-IDF (Local Model) and OpenAI Embeddings for intent matching. The library allows users to dynamically add or remove intents, cache frequent queries, and choose between algorithms for better flexibility.

📦 Installation

npm install intent-classifier

🔧 Usage

1️⃣ Basic Example - TF-IDF Based Matching

const IntentClassifier = require("intent-classifier");

const classifier = new IntentClassifier({
    useOpenAI: false, // Use TF-IDF-based classification
    dataset: [
        { text: "How do I book a flight?", intent: "book_flight" },
        { text: "I need to reserve a ticket", intent: "book_flight" },
        { text: "What's the weather like?", intent: "weather_query" }
    ]
});

async function run() {
    const result = await classifier.classify("How do I book a flight?");
    console.log(result); // { intent: 'book_flight', confidence: '98.42%' }
}

run();

2️⃣ OpenAI Embeddings-Based Matching

const classifier = new IntentClassifier({
    apiKey: "your-openai-api-key", // Required for OpenAI embeddings
    useOpenAI: true,
    dataset: [
        { text: "How do I reset my password?", intent: "password_reset" },
        { text: "I want to create an account", intent: "account_creation" }
    ]
});

async function run() {
    await classifier.generateEmbeddings(); // Generate embeddings for dataset
    const result = await classifier.classify("How do I reset my password?");
    console.log(result); // { intent: 'password_reset', confidence: '99.21%' }
}

run();

⚙️ Configuration Options

| Option | Type | Default | Description | |-----------------|----------|---------|-------------| | useOpenAI | Boolean | false | Use OpenAI embeddings for classification | | apiKey | String | null | OpenAI API key (Required if useOpenAI is true) | | threshold | Number | 0.50 | Confidence threshold for intent classification | | batchSize | Number | 50 | Batch size for OpenAI embeddings | | embeddingsFile| String | ./embeddings.json | Path to store OpenAI-generated embeddings |

📌 Methods

🔍 classify(query: string): Promise<{ intent: string, confidence: string }>

Classifies a user query into an intent.

🔍 addIntent(text: string, intent: string): Promise<void>

Dynamically adds a new intent to the dataset and updates embeddings if using OpenAI.

🔍 removeIntent(intent: string): Promise<void>

Removes an intent from the dataset and updates embeddings if using OpenAI.

🔍 generateEmbeddings(): Promise<void>

Generates and saves embeddings for the dataset (Only for OpenAI mode).

🚀 Additional Features

• Garbage Query Detection: Uses heuristics and OpenAI to filter out nonsensical queries.
• Jaccard Similarity Filtering: Prevents force-matching by checking word similarity.
• Query Caching: Avoids redundant API calls for repeated queries.

🛠 Contributing

Contributions are welcome! Feel free to open an issue or submit a pull request.

📝 License

This project is licensed under the MIT License.