Robust evaluation of binary collaborative recommendation under profile injection attack

Abstract

Recommender systems are being improved by every means to be more accurate, more robust, and faster. Collaborative filtering is the mainstream type of recommendation algorithms, and its core is calculating the similarity between users or items based on ratings. Researchers recently found that the binary similarity based solely on who-rated-what rather than actual ratings output more accurate recommendation. We, from robust perspective, evaluated the binary collaborative filtering under multiple types of profile injection attacks on large dataset. Experimental results show binary collaborative filtering is more robust than actual ratings based collaborative filtering in all situations.