The Micro-scale Adversarial Detection Algorithm for Overlapping Communities Based on Genetic Algorithms

Currently, most methods for adversarial community detection focus on macroscopic scales based on overall community structures and on mesoscopic scales centered around target communities. These methods primarily involve rewiring (deleting old links and adding new ones) non-overlapping communities, with limited research on micro-scale adversarial tactics within overlapping communities at target user nodes. To address this gap, this paper takes a micro-scale approach, we propose the Genetic algorithm-based Micro-scale adversarial detection method for Overlapping communities (GOM). This method employs the addition of only fake user nodes and links to transform single-community user nodes into multi-community user nodes at minimal cost, aiming to conceal information about the target nodes. The paper introduces metrics for evaluating the effectiveness of hiding target nodes and assessing the strengths and weaknesses of adversarial strategies, with the cost of reconstructing data structures considered as one of the evaluation metric factors. The paper tests multiple community detection algorithms on various real network datasets using evaluation metrics and compares them with four adversarial overlapping community methods. The performance metrics of the GOM method are higher than those of traditional baseline methods.