Spoilers Ahead -- Personalized Web Filtering

摘要

Unwanted content on web pages can take many forms, be it ads, malicious code, pointless clutter, or specific topics that the user does not want to read about (yet). Unlike most other work, we focus on the latter. The user can define terms based on which we prevent the disclosure of undesired information (e.g., the latest sports result) and warn the user before it is revealed. The user can decide if and when the filtered elements should be displayed. We define this formally as the node removal problem and show its equivalence to the NP-hard knapsack problem. We developed a proof of concept Firefox extension to filter web pages based on user defined terms and our heuristic. Our evaluation shows that we correctly distinguish between wanted and unwanted content in approximately 9 out of 10 cases.