Cooperative Behavior Detection In Social Media Trending Topics

In recent years, the rapidly developing online social media is an important source of information in people’s life. However, as the most attractive content of social media, trending topics have become the main target for malicious behaviors, such as advertising, viral marketing, rumours and public attacks. They have a lot in common that a number of accounts promote the target topic cooperatively. This kind of behavior can worsen the cyberspace, reduce the social media content quality, and may even cause social harm and serious legal consequences. In this paper, we study the massive social media data, analyze the mechanism of cooperative behavior in trending topics. We formalize the problem and propose a framework for analyzing cooperative behaviors in trending topics, design and implement a distributed online detecting method. Experimental results show that the framework is capable of finding cooperative behaviors in trending topics, and the online detecting method archives good effect and high system throughput.The contributions of this paper are threefold:● We formalize the problem of cooperative behavior detection in trending topics. We formally define the trending topics and cooperative behavior in our data model, present criteria for evaluation. The general idea of solving this problem is given, i.e., trend discovery by calculating the cohesive subgroups, and building up statistical model for cooperative behavior analysis.● We present a framework for cooperative behavior detection in trending topics. This framework includes two parts, i.e., candidate trend discovery and coopera-tive behavior analysis. We propose an edge based algorithm for finding cohesive subgroups as trending topics, and community association distance for cooperative behavior analysis. Experimental results show that the framework can deal with mas-sive social media data, pick out trends and analyze the cooperative behaviors.● We present a distributed online detecting method. We design distributed algo-rithms for message graph construction and cohesive subgroups computing, imple-ment them on Spark, and train AdaBoost model for online classification. Exper-imental results show that our method can achieve high detecting accuracy, high throughput and low response delay, which meets the needs of online applications.Research on cooperative behaviors in trending topics enable us to gain a deeper un-derstanding of its kinds, mechanism and motivation, with behavior science and sociology significance. Analysis of cooperative relations can be used to improve the social media anti-spam system, shielding spam users and messages. The online detection of cooper-ative behavior also has important practical significance, which can realize and eliminate threats timely, provide users high quality content and enhance the user’s experience.