Comprehensive decomposition optimization method for locating key sets of commenters spreading conspiracy theory in complex social networks

Abstract

With the power of social media being harnessed to coordinate events and revolutions across the globe, it is important to identify the key sets of individuals that have the power to mobilize crowds. These key sets have higher resources at their disposal and can regulate the flow of information in social networks. They can maximize information spread and influence/manipulate crowds when they are coordinating. But due to the inherent drawbacks in node-based and network-based community detection algorithms, neither of these types of algorithms can be used to detect/identify these key sets. In this study, we present a bi-level max-max optimization approach to identify these key sets, where the degree centrality is used to identify individuals’ influence at the commenter-level, while the network-level is designed to evaluate the spectral modularity values. We also present a set of evaluation metrics that can be used to rank these key sets for an in-depth investigation. We demonstrated the efficacy of the proposed model by identifying key sets hidden in a YouTube network spreading fake news about the conflict in South China Sea. The network consisted of 47,265 comments, 8477 commenters, and 5095 videos. A co-commenter network was constructed, where two commenters were linked together if they comment on same video. The proposed model efficiently identified key sets of commenters spread information to the whole network to manipulate YouTube’s recommendation and search algorithm to increase the information dissemination. Moreover, the projected approach could identify sets of commenters that were key connectors to multiple groups, high influence across the network, higher interactions, and reachability than other regular communities. Besides, the Girvan–Newman modularity method, the depth-first search method, and text analysis was applied to validate the outcomes, categorize the identified key sets, and monitor the commenters’ behaviors and information spread strategies in the network. In addition, the model considered a multi-criteria problem to rank these key sets of commenters based on the small real-world networks’ features.