AdGraph: A Graph-Based Approach to Ad and Tracker Blocking(I. INTRODUCTION)

I. INTRODUCTION

Abstract

The need for online content blocking continues to grow.

• Studies show that blocking ads and tracking resources can lead to:
  • Improved performance
  • Enhanced privacy
  • Stronger security
  • Better user experience
• Browser vendors are increasingly integrating content blocking features directly into browsers.

Limitations of existing content blocking tools

• Traditional tools rely on URL patterns or JavaScript behavior/code structure to block suspicious content.
  • Limitations:
    • Can be bypassed via domain generation algorithms (DGAs)
    • Can be evaded using trusted domains (e.g., first-party or CDN proxies)
    • Can be obfuscated through JavaScript code manipulation

Limitations of alternative approaches

• Previous research has proposed filter lists, predefined heuristics, and machine learning based on network/code analysis, but these are often incomplete or vulnerable to simple evasion techniques.

---

ADGRAPH Contributions

What is ADGRAPH?

• ADGRAPH is a novel machine learning-based approach that detects and blocks advertising and tracking resources by constructing a graph based on HTML structure, JavaScript behavior, and network request interactions.
• It provides a context-rich blocking strategy:
  • Takes into account both past and current context of network requests to detect resources missed by conventional methods
  • More robust against single-feature-based evasion techniques

Key Features

1. Supports both online (real-time) and offline usage
   • Online: Blocks ads and trackers in real time within the browser
   • Offline: Assists in generating and maintaining filter lists
2. High performance
   • Demonstrates faster page load times compared to Adblock Plus

Summary of ADGRAPH’s Contributions

1. Graph-based machine learning approach
   • Utilizes HTML structure, JavaScript behavior, and network requests to identify ad/tracking resources
2. Large-scale evaluation
   • Reproduced filter list labels with 95.33% accuracy across the Alexa Top 10,000 websites
   • Successfully identified resources missed by filter lists and distinguished overblocked legitimate resources
3. Efficient implementation
   • Modified Chromium’s Blink and V8 engines to monitor document behavior
   • Outperforms Adblock Plus, and is even faster than vanilla Chromium on 42% of tested websites
4. Site breakage evaluation
   • Causes comparable site breakage to filter lists (15.0% vs. 11.4%)
   • Induces fewer severe breakages (5.9% vs. 6.4%)

---

Paper Structure

1. Section II: Reviews existing work on ad and tracker blocking and its limitations
2. Section III: Describes the design and implementation of ADGRAPH
3. Section IV: Evaluates ADGRAPH in terms of blocking accuracy, performance, and site breakage
4. Section V: Discusses limitations, improvements, and potential for offline use
5. Section VI: Conclusion