Research On Spammer And Rumor Detection Method In Weibo Based On Machine Learning

In recent years,online social networks have risen rapidly,and the emerging social media represented by Weibo have exerted a profound influence on people's lives by its amazing spread speed and spread breadth of communication.On the Weibo platform,the user is the main body,and the content of Weibo is an important carrier of information dissemination.However,in the last few years,the problem of spammers and rumors on the Weibo platform has become increasingly serious,which not only affects the data availability and security of the platform,but also causes problems such as unhealthy development of the platform and poor user experience.Therefore,it is of great practical significance to design effective methods to detect spammers and rumors.This paper conducts an in-depth research on the two types of problems of spammers and rumors on the Weibo platform.The specific research contents are as follows:1.In order to detect spammers more effectively in Weibo,an approach based on multi-view fusion was proposed.First,a user representation strategy for integrating multi-view information was designed to characterize users from 3 views,namely user behavior,social relationship and text content.In view of the deficiencies that the existing approaches do not fully consider the user's fans and user's environment in social networks,new features such as fan ratio,fan average bidirectional connection rate,community-based bidirectional connection rate,community-based cluster coefficient,etc.were introduced.A linear weighting fusion was carried out based on the classification results from each view.The optimal fusion coefficient was obtained by minimizing the approximate error,and then the final classification result was obtained.The test result on the real data from Weibo show that this approach can not only effectively detect spammers,with significant improvement in precision and F1-sorce,but also exhibits greater stability especially when processing unbalanced data.2.For the rumor problem on the Weibo platform,the current detection models mostly apply the traditional feature engineering method,which is data dependent,requires the researchers to have a strong background knowledge,and is easy to cause the problem of feature missing or redundancy.This paper proposes a rumor detection model based on Bert-CNN.Events as the research object,characterization of the Weibo text contained in the event using the BERT,and a feature transformation algorithm is designed to obtain the event level encoding.The encoded results are taken as the input of the CNN.Through convolutional neural network hidden layer's learn to mine the deep features of the event,and finally get the final classification result through the sigmoid function.Experimental analysis shows that the proposed method has a greater improvement in detection accuracy than the previous methods,and has better early detection capability.