Sybil Detection using Hybrid Approach

Abstract

Online social networks (OSNs) are getting viral day by day. Everyone is getting habitual to OSNs. It is a source of entertainment but it is also a source of threat to its potential users as in form of fake profiles, spam etc. Security problems regarding fake profiles and spam on OSNs is genuine issue. This study targets to create methodology that will resolve these kinds of issues. This work is story of propagating belief, calculating probabilities (prior or posterior) iteratively on nodes in a graph, and after that final reputation scores will classify nodes in nodes of OSN by assigning different probabilities/weights to different nodes simultaneously keeping in mind the nature of nodes. Graph theory methods and machine learning algorithms or hybrid approaches will be used. Existing methods simply set a fix weight to all edges assign, prior probability using random walk (RW) or loopy belief propagation (LBP), after that they classify dataset randomly or manually in prior stages. In this work, a novel collective classification method is addressing this long-standing challenge. Machine learning and Sybil detection algorithms will be used for pre-processing of data labels, then Random Walk and Loopy Belief Propagation (LBP) will be used to assign weights and beliefs, which will lead toward the goals about the final reputation scores. Through learning edge weights, this work propagates the reputation scores using propagating methods to achieve better accuracy rate. Partial knowledge is used to make inferences on millions of vertices and corresponding edges. Random Walk and Loopy Belief Propagation is used in applications including social network analysis and bots’ detection. Various linearly running algorithms are used to find out marginal probabilities of vertices. As there are billions of vertices in real world OSNs, which will cause lagging so this issue should be resolved. Fitting this large network in one memory is a problem. Therefore, multithreads would be used, which will lower the burden and run threads simultaneously. Distributed or parallel computation are used to resolve these problems.