Science Diliman (Dec 2011)
TULUNGAN: A Consensus-Independent Reputation System for Collaborative Web Filtering Systems
Abstract
Web filtering systems allow or prohibit access to websites based on categories (e.g., pornography,violence, sports, etc.). Categorization of websites can be done automatically or manually. Automaticcategorization is prone to under- and over-blocking. On the other hand, manual approach is typicallyperformed by a limited number of people making it not scalable.Collaborative web filtering systems, a variation of manual categorization, allow anyone to categorizewebsites in order to determine which domain these sites belong (e.g., pornography, violence, sports,etc.). This attempts to solve the scalability issue of the typical manual method.The approach offered by collaborative web filtering relies heavily on the contribution of users in orderto make the system scalable and less prone to errors. However, its success is greatly dependent on usercooperation. To promote cooperation, reputation system can be used in web filtering.A previous study called Rater-Rating promotes cooperation and explores the use of a user-drivenreputation system that measures both the contributor and rater reputation of users of a collaborative websystem. However, Rater-Rating is consensus dependent. If the number of malicious users are more thantheir good counterparts, the reputation system can be defeated. In other words, the system canmistakenly give malicious users a high reputation value.This study discusses a reputation system called Tulungan that is consensus-independent. It can detectthe presence of malicious users even if the number of their good counterparts are fewer. A simulationresult that compares the effectiveness of Tulungan relative to Rater-Rating is presented in this paper.The simulation shows that Tulungan is still effective even with 25% good users while Rater-Ratingrequires at least 50% good users to be effective.
