Co-Detection Of Crowdturfing Microblogs And Spammers

In recent years,with the rapid development of Internet technology,online social network has become a significant means for people to get real time information.However,the convenience of disseminating information also makes it the tool for spammers to intersperse ads and fraud contents,which has severely negative impact on network environments.Besides,the rise of crowdsourcing services offers new ways for online collaboration and results in the emergence of crowdsourcing platforms.Large numbers of crowdturfing tasks on these platforms tempt many crowdsourcing users to participate into marketing promotion campaigns in social network sites just as spammers.Different from conventional spammer accounts controlled by scripts,the owners of these crowdsourcing participants are usually real people,showing more similarities with normal users.Therefore,the contents they post are more deceptive and bring new challenges for spam detection.In this thesis crowdturfing in Sina Weibo is studied.Existed researches lacked deep-level mining of crowdsourcing mechanism and ignored the close relation between crowdturfing spammers and microblogs.A co-detection model is proposed to solve these problems.Starting from the aspect of both users and microblogs,the model analyzes their abnormal features in crowdturfing campaigns and takes the topological relation of following and retweeting into consideration.The main work includes:Firstly,topological models of following relation between users as well as retweeting relation between users and microblogs are constructed.Based on the ideas of label propagation and optimization,corresponding optimal target functions are given.On this basis,we propose crowdturfing microblogs and spammers co-detecting model,which transforms the detection problem into a constrained optimization problem.Secondly,inspired by the specific mechanism of crowdturfing,feature mining is done from the aspect of user characteristics,behavior,comments,etc.And those features are treated as prior knowledge for our co-detecting model.After that,based on the ideas of alternate iteration,the Crowdturfing Microblogs and Spammers Co-detection Algorithm(CMSCA)is proposed to solve the model.Finally,experiments are carried out on real crowdturfing dataset to verify the effectiveness of our algorithm.We can conclude from the experiment results that compared to the algorithms for contrast,our CMSCA algorithm gets better detection performance.There is also an obvious enhancement of the utility of abnormal features.In addition,it is proved that the ideas of codetection takes advantage of the relation between crowdturfing spammers and microblogs,from which the performance of detection of both sides are benefited.The research of crowdturfing in social medium can help to improve the performance of spam detection of those websites and cut the wastage of network bandwith caused by overflow of ads and fraud contents.It will also contribute to the supervision of public opinion and will create a better network environment for users.