Tracking Detector

Tracking Detector is a proof of concept for real time web tracker detection. The extension has an underlying DNN model which can identify tracking from non-tracking web requests.

Model

The model is a tensorflowjs sequential dnna and expects a feature vector of dimension (1, 204) and returns a single number between 0 and 1. One being a tracker and 0 being a non-tracker.

Feature Extraction

The feature vector is build in real time and encodes the url of the web requests as well as other information.

// Part of FeatureExtractor.js
const generateURLEncoding = (url) => {
        let encoding = [];
        for (let i = 0; i< url.length; i++) {
            encoding.push((url.charCodeAt(i) % 89) +1 )
        }
        if (encoding.length < 200) {
            encoding = new Array(200 - encoding.length).fill(0).concat(encoding);
        } else if (encoding.length > 200) {
            encoding.splice(0, encoding.length - 200);
        }
        return encoding;
    }
return {
    encode(request) {
        let features = [];
        features = features.concat(generateURLEncoding(request.url));
        // features.push(request.frameId > 200000 ? 199999 : (request.frameId+1));
        features.push(frameTypes.indexOf(request.frameType)+1);
        features.push((methods.indexOf(request.method)+1));
        features.push((types.indexOf(request.type)+1));
        features.push((request.requestHeaders.filter(x => x.name == "Referer").length == 1 ? 1 : 0));
        const tensor = tf.reshape(tf.tensor(features), [1, 204]);
        return tensor
    }
}

The url gets transformed to a sequence of integer values between 1 and 90. The model only uses 200 characters of the url and adds padding when the url is to short or deletes the starting characters.

Additional features

frameType

The frameType gets transformed to an integer value between 1 and 3.

// All possible frameTypes
const frameTypes = [
    "outermost_frame",
    "fenced_frame",
    "sub_frame",
]

httpMethod

The httpMethod gets transformed to an integer value between 1 and 8.

// All possible methods
const methods = [
        "GET",
        "POST",
        "OPTIONS",
        "HEAD",
        "PUT",
        "DELETE",
        "SEARCH",
        "PATCH",
]

httpMethod

The requestType gets transformed to an integer value between 1 and 12.

// All possible requesttypes
const types = [
        "xmlhttprequest",
        "image",
        "font",
        "script",
        "stylesheet",
        "ping",
        "sub_frame",
        "other",
        "main_frame",
        "csp_report",
        "object",
        "media",
]

Sample Encoding

0,0,...,16,28,28,24,27,59,48,...,1,1,2,1

This vector includes a padding at the beginning the encoded url and the additional attributes.

Build

Clone the repository.

git clone https://github.com/HenrySchwerdt/tracking_detector.git

Install Dependencies

npm install

Build Project

npm run build

Installation

You can directly install the extension from the public folder when Chrome developer options are enabled.