Adaptive Crawling in Online Social Networks: a Matrix Factorization Based Approach

Privacy issues in Online Social Networks (OSNs) are severer when more and more users are sharing their candid personal information and friendships online. One simple attack aims at private user data is to use socialbots to befriend the users and crawl data from users who are tricked to accept the friend requests. This attack is shown to be effective and hence it is important to understand the friending strategies of this attack. The existing works on the friending strategies focus on the theoretical side to provide guarantees for near-optimal strategies and assume that the attackers know the probabilistic network topology as well as the friend request acceptance rates. In this work, we instead focus on a practical problem: the attackers usually have very restricted knowledge of the OSNs before the attack. We aim at solving the Adaptive Crawling in OSNs (ACOSN) problem, in which the attackers make decisions using only the crawled data. We design our strategy with the help of a matrix factorization based link prediction approach that can adaptively update the link existence probabilities using current knowledge. The experiment results show that our approach is effective in actual OSN data sets.