Detection And Difference Research On Social Bots

With the development of automation programs,automated accounts,namely social robots,have also become active on social platforms.Although social robots can provide users with efficient and convenient services,the malicious use of some social robots,such as the dissemination of false information and manipulation of public opinion,has also brought serious negative impacts on social networks and users.Therefore,this paper focuses on the detection of social robots and conducts in-depth research on the Twitter platform to provide security for the healthy development of social networks.The research content of this paper is as follows:First of all,due to the update iteration of social robots,the differential features are more hidden,and the method of improving the detection performance by increasing the number of features will undoubtedly increase the complexity of feature engineering.In addition,existing work ignores the correspondence between account metadata and tweet metadata in social bots.Therefore,this paper proposes a social robot detection method based on BiLSTM and attention mechanism.This method starts from the perspective of tweets and uses the corresponding relationship between account metadata and tweet metadata as auxiliary features for social robot detection.Specifically,we first use the BiLSTM neural network to automatically capture contextual information,learn complex language structures and expressive features in tweets,and thus extract deep semantic features of tweets.Subsequently,the data features extracted from tweet metadata and account metadata are combined with tweet semantic features to reveal typical patterns and abnormal behaviors of social robots,and an attention mechanism is introduced to obtain key information that affects classification results,thereby improving social interaction.Precision and reliability of robot detection.Experimental results show that the proposed method can significantly reduce the cost of social bot detection and outperform other detection methods using only a single tweet and related information.Secondly,although the existing work has detected social robots on the Twitter platform,it has not deeply analyzed the differences between social robots and real users and between different social robots.In addition,most of the existing work is based on public data sets with relatively backward timeliness,resulting in a gap in the analysis of the abnormal behavior of new social robots during the Russo-Ukrainian War.Therefore,this paper first designs a Twitter topic data collection framework that combines crawler technology and API technology,and collects relevant data on the Twitter platform within 2 months since the outbreak of the Russo-Ukrainian War.Secondly,this paper uses the detection method proposed above to detect the RussianUkrainian War dataset,and analyzes the differences between social robots and real users from the two dimensions of account information and tweet data.Finally,this paper proposes a BERT+TextCNN-based social robot classification method to analyze the differences between different types of social robots in the Russian-Ukrainian War dataset.Different from detection methods that only distinguish between real users and social bots,this method classifies social bots at a more fine-grained level based on the behavioral differences between them based on the detection results of social bots.The results of differential analysis help researchers to defend against social robots with malicious behaviors.