SpamWarden.js

Lightweight, client-side JavaScript library for detecting spam and sentence hijacking in real-time. Trained on the model from RedSocs/spam-labeler, bundled for zero-dependency browser usage.

Quick Start

Install

# npm
npm install @_redsocs/spam-warden

# Or download from CDN / GitHub releases

In the Browser

<!-- Option 1: From RedSocs CDN -->
<script src="https://redsocs.com/js/spam-warden.js"></script>

<!-- Option 2: Self-hosted -->
<script src="dist/spamwarden.min.js"></script>

<script>
  const result = window.spamwarden.spamcheck("สมัครสมาชิกวันนี้ รับโบนัส ฟรี!");
  console.log(result.isSpam);   // true
  console.log(result.prob);     // 1.0
  console.log(result.version);  // "v0.69"
</script>

As ES Module

import SpamWarden from './dist/spamwarden.min.js';
SpamWarden.spamcheck("Welcome bonus! Deposit now");

Quick Boolean Check

if (spamwarden.isSpam(userInput)) {
  // block or flag
}

In Node.js

const spamwarden = require("./dist/spamwarden.min.js");
const r = spamwarden.spamcheck("Welcome bonus! Deposit now get 200% match");
console.log(r.isSpam); // true

API

spamwarden.spamcheck(text) → object

| Field | Type | Description | |-------|------|-------------| | isSpam | boolean | true if detected as spam | | prob | number | Spam probability (0.0–1.0) | | reason | string? | Present if hard-rule triggered: "currency_symbol" or "spam_link" | | version | string | Model version (e.g., "v0.69") |

spamwarden.isSpam(text) → boolean

Convenience wrapper — returns only the boolean result.

spamwarden.version → string

Current model version string.

Build

# 1. Copy model from spam-labeler
cp ../spam-labeler/extension/model.json .

# 2. Build (bundles model into JS)
node build.js
# or: ./build.sh

Output:

| File | Size | |------|------| | dist/spamwarden.js | 3.5 MB (uncompressed) | | dist/spamwarden.min.js | 61 KB (minified) | | dist/spamwarden.min.js (gzipped) | 27 KB |

Optional: Better Minification

npm install terser
node build.js   # now uses terser instead of simple minification

How It Works

User posts text
    ↓
spamwarden.spamcheck(text)
    ↓
Hard rules check (currency symbols, spam links)
    ↓
Vectorizer: whitespace tokens + trigrams + quadgrams
    ↓
Bernoulli Naive Bayes prediction (class 0 = safe, 1 = spam)
    ↓
Softmax → probability
    ↓
{ isSpam, prob, version }

Model

| Property | Value | |----------|-------| | Origin | RedSocs/spam-labeler (Rust, Bernoulli NB) | | Features | ~63,000 tokens (whitespace + trigrams + quadgrams) | | Version | v0.69 (680 training samples) | | Hard Rules | Currency symbols ($€£฿) → auto-spam; Spam links (line.me, @line, lin.ee) → auto-spam |

Train Your Own Model

The model in this repo was trained by RedSocs/spam-labeler. To customize for your own use case:

# 1. Clone the training repo
git clone https://github.com/RedSocs/spam-labeler.git

# 2. Add your own training data
cp your-spam.txt spam-labeler/data/spam.txt
cp your-safe.txt spam-labeler/data/safe.txt

# 3. Retrain and export
cd spam-labeler
cargo run --release --bin export_model
cp extension/model.json ../spam-warden/model.json

# 4. Rebuild SpamWarden
cd ../spam-warden
node build.js

See the spam-labeler README for the full training pipeline.

Privacy

All processing happens in-memory in the browser. No data is sent to any server.

Related

• RedSocs/spam-labeler — Rust-based training pipeline, TUI app, and Firefox extension for collecting and training the spam detection model.

Project Structure

spam-warden/
├── src/
│   └── spamwarden.js    # Library source (MODEL_DATA_PLACEHOLDER)
├── dist/
│   ├── spamwarden.js    # Bundled (model inlined, ~3.5 MB)
│   └── spamwarden.min.js # Minified for production (~61 KB, 27 KB gzipped)
├── model.json           # Trained model from spam-labeler
├── build.js             # Node.js build script
├── build.sh             # Shell wrapper
└── README.md            # This file