AuraClassify

AuraClassify is a powerful content moderation and classification system built on TensorFlow.js, utilizing the Universal Sentence Encoder for text analysis.

Features

• Content moderation and classification
• Multi-category support with subcategories
• Confidence scoring and sentiment analysis
• Detailed analysis reports
• Easy to train and use
• Supports both browser and Node.js environments

Installation

npm install auraclassify

Quick Start

const AuraClassify = require('auraclassify');

// Initialize classifier
const classifier = new AuraClassify({
    backend: "tfjs" // or "tfjs-node" for Node.js backend
});

// Train the model
await classifier.train({
    dataset: trainingData,
    log: true,
    batchSize: 4
});

// Classify text
const result = await classifier.classify("Text to analyze");

Training Data Format

Training data should be an array of objects with input and output properties:

const trainingData = [
    {
        input: "Example text content",
        output: "category" // or "category/subcategory"
    }
];

Supported Categories

• safe: Safe content
• sexual: Adult content
• harassment: Harassment content
• hate: Hate speech
• illicit: Illegal content
• self-harm: Self-harm content
• violence: Violent content

Each category can have subcategories (e.g., "violence/threatening", "self-harm/instructions")

API Reference

Constructor

const classifier = new AuraClassify({ backend = "tfjs" });

Methods

train(options)

await classifier.train({
    dataset: trainingData,
    log: true,
    batchSize: 4
});

classify(text)

const result = await classifier.classify("Text to analyze");

save(path)

await classifier.save("path/to/model.json");

load(path)

await classifier.load("path/to/model.json");

How It Works

AuraClassify uses the Universal Sentence Encoder to convert text into high-dimensional vectors (embeddings). These embeddings capture semantic meaning, allowing the system to understand context and nuance in text.

The classification process involves:

1. Text embedding generation
2. Similarity comparison with trained examples
3. Category and subcategory detection
4. Confidence scoring
5. Detailed analysis generation

Example Output

{
    analysis: {
        input: {
            text: "Original text",
            length: 12,
            wordCount: 2
        },
        result: {
            label: "category/subcategory",
            confidence: 0.85,
            confidenceLevel: "HIGH"
        },
        // ... additional analysis data
    },
    summary: {
        decision: "CATEGORY (HIGH confidence level)",
        confidence: 0.85,
        status: "RELIABLE"
    }
}

License

Apache License 2.0

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

Support

For issues and feature requests, please use the GitHub issues page.