Study On Social Bot Detection Based On Sentiment And Network Features

With the rapid development of the Internet,there are more and more social bots controlled by automatic programs in major social networks,and some of them can even mimic human behavioral activities to achieve the effect of faking the real thing.However,some malicious social bots have seriously affected the security of social networks and the stability of social life by spamming phishing websites,steering public opinion,stealing users’ privacy and creating rumors.The detection of malicious social bots has become a very important direction of network security.The current detection methods for malicious social bots are basically based on account metadata analysis,and there is less research on sentiment analysis and social relationships.In order to improve the detection result of social bots,this paper proposes a Bidirectional Long Short-Term Memory model based on attention mechanism(Bi-LSTM+ Attention)to perform sentiment analysis on the text content of social bot accounts,and obtain new sentiment features through calculations,feature fusion of the obtained sentiment features and the extracted metadata features.Then the fusion features are supervised by five machine learning models and convolutional neural network(CNN)for detection and classification.Through comparative experiments with metadata features,the detection results of all detection models have been improved to a certain extent,and the detection accuracy has reached more than 97%.At the same time,in order to explore the influence of relational network relations on the detection effect of social bots,this paper proposes a graph convolution(GCN)model based on its detection process.A graph is constructed through the relationship network between social bots,and the constructed graph adjacency matrix and metadata features are input into the GCN model for classification.Finally,good results have been achieved,which proves the feasibility and effectiveness of the method.The above experimental data show that sentiment features and network features play a positive role in the detection of social bots.In the future,when social bots are more anthropomorphic,social network and sentiment analysis will provide better ideas for social bot detection.It can provide better protection for social network security.