Leveraging Crowdsourcing for Efficient Malicious Users Detection in Large-Scale Social Networks

The past few years have witnessed the dramatic popularity of large-scale social networks where malicious nodes detection is one of the fundamental problems. Most existing works focus on actively detecting malicious nodes by verifying signal correlation or behavior consistency. It may not work well in large-scale social networks since the number of users is extremely large and the difference between normal users and malicious users is inconspicuous. In this paper, we propose a novel approach that leverages the power of users to perform the detection task. We design incentive mechanisms to encourage the participation of users under two scenarios: 1) full information and 2) partial information. In full information scenario, we design a specific incentive scheme for users according to their preferences, which can provide the desirable detection result and minimize overall cost. In partial information scenario, assuming that we only have statistical information about users, we first transform the incentive mechanism design to an optimization problem, and then design the optimal incentive scheme under different system parameters by solving the optimization problem. We perform extensive simulations to validate the analysis and demonstrate the impact of system factors on the overall cost.