Design And Experimental Analysis Of Social Bot Detection Algorithms In Online Social Networks

Social bots are individual entities or collections of entities controlled by humans or algorithms,designed to mimic human account behavior for specific purposes.Over the past decade,a significant number of bots have emerged on various social platforms.These social bots engage in activities such as fabricating and spreading false information,as well as influencing public sentiment.These behaviors not only harm the legitimate rights of real users but also have a negative impact on the healthy development of online social systems and social stability.Therefore,detecting these social bot accounts is of great significance.The thesis takes social bot behavior analysis as its starting point and extracts an original feature set.In order to reduce feature redundancy and improve algorithm efficiency,a feature selection method based on PCA-A is proposed.An efficiency-harmony metric is set to evaluate the performance of this method.Experimental analysis demonstrates that the random forest algorithm performs the best when the selected feature set accounts for 90% of the original feature set.Additionally,the thesis presents an improved random forest algorithm that incorporates feature selection and parameter tuning using cross-validation.The performance of the Naive Bayes algorithm,decision tree algorithm,k-means algorithm,and the original random forest algorithm is compared separately for user features,tweet features,and hybrid features.Finally,the performance of the improved random forest algorithm is compared with that of the original random forest algorithm.The experimental results indicate that,in the presence of user features,tweet features,and a combination of both,the original random forest algorithm outperforms the other three algorithms in terms of accuracy,precision,and recall for social bot detection tasks.Furthermore,the improved random forest algorithm shows even better performance compared to the original random forest algorithm.Overall,the combined approach utilizing user features,tweet features,and the improved random forest algorithm demonstrates promising results in the identification of social bots.