Structure And Content-Based Spammer Detection In Social Networks

With the continuous development of the Internet, information dissemination based on social network is becoming far more in-depth and extensive. However, organized network spammers appearing recent years lead to the prevalence of rumors and the serious rise of cheating, which make huge damage to social and economic environment and influence the fundamental of social network, will finally harm the development of social network. As a result, detection of network spammers is indeed an imminent task. Traditional ways of detecting spammers, in the huge social network, are based on a single feature, which leave room for improvement. How to recognize spammer groups in social network and improve its efficiency and precision is therefore a significant research point.Network spammer groups, as organizations with certain tasks, possess abnormal characteristics among their membership structure. For this reason, we propose our method to recognize network spammers based on combined structure and content features. We can detect spammer groups by figuring out their structure features in the social network. After considering content features they spread, we can are able to identify spammers by comprehensive analysis. Our work is as follows:First, we distinguish structure features of spammers in social network. By recognizing temporal characters when garbage information appears and constructing forwarding relation networks using forwarding records among users to find overlapping community structures that are of strong propagating ability, we can make preliminary identification of network spammers.Second, we extract features among contents that users spread. By measuring content characters users send and those of certain garbage information, we are able to make judgment whether garbage information a user has ever sent.Finally, we recognize spammer groups considering both structure and content features. Based on overlapping communities that have been recognized, we measure similarity between contents in these communities and certain garbage information, we are capable of recognizing communities that have spread garbage information more than a certain times, and we confirm them as network spammers.Our method has global characteristic in that the detection method it uses is based on the overall structure of network spammers. After comparison with other methods, experimental results on Sina Weibo data set have proven our work with higher efficiency and feasibility. Related technical achievements may help purify the environment of social network and will have huge application prospect.