AI AdBlocker (for Android) 作者: eye3

No user data is collected, stored, or transmitted. No remote endpoints. All ML/WASM runs entirely locally.
This add-on is a starter AI Ad Block extension.
• The uploaded .zip is the Firefox-compatible build (no rules/ directory, uses background.scripts instead of service_worker).
• The uploaded source .zip contains the full, human-readable source code (TypeScript, scripts, manifests, build files) without dist/ or node_modules/.
• No remote code execution, dynamic code generation, or obfuscated code is used.
• Can detect hidden ads, sponsored labels, promoted posts, native ads that rule-based systems miss.
• Can adapt better if you re-train the model with new data.
Example AI Features You Can Add DOM / Heuristic Classifier:
Train a lightweight ML model on HTML snippets (features: tag type, attributes, text like “Sponsored”).
Content script grabs candidate nodes → runs model → hide if classified as ad.
Vision-based Ad Detection:
Use a small CNN (e.g., MobileNet/ONNX quantized) to check if an <img> looks like a banner ad.
Useful for “image-only” ads where markup doesn’t give them away.
Hybrid (most practical):
Use heuristics to filter likely candidates (divs with fixed size, suspicious classes, “sponsored” text).
Use ML to confirm → avoid false positives.
Future-proofing: when advertisers obfuscate HTML/CSS, rules break → but your ML model still generalizes.
Privacy-preserving: everything runs locally in the browser; no need to send page data to servers.
Research value: positions your extension as “next-gen” ad blocker, different from commodity ones.