Research And Implementation Of Social Robot Detection Technology Based On Improved Conditional Generation Adversarial Network

With the rapid development of the Internet world,online social networks have become an indispensable part of people’s daily lives.A large number of social robots controlled by automated programs have emerged on online social networks.These social robots simulate the browsing behavior and activity content of normal humans,and establish a trust relationship with normal humans,so as to achieve the purpose of launching social engineering attacks,which poses a non-negligible harm to cyberspace security.It is detected and deleted in online social networks malicious social robots have become a key area of focus for industry and academia.There are significant differences in the number of social robots and normal human users in the real environment.Machine learning-based robot detection methods will cause classifier bias due to the imbalance in the number of samples in different categories,and the detection rate of a few samples is low.In order to ensure the accuracy of social robot detection,this paper proposes an improved CGAN(Conditional Generative Adversarial Networks)to solve the problem of data imbalance.At the same time,a density peak clustering algorithm based on Gaussian kernel is used to generate condition variables.The social robot detection system puts theoretical technology into practice.The main research results and innovations of this article are as follows:(1)A Gaussian Kernel Density Peak Clustering Algorithm is proposed The original DPCA used Euclidean distance to calculate the distance of the data points.When the sample space is linear and indivisible,the original DPCA is misclassified.The Gaussian kernel distance is used to improve the original DPCA,and the distance is measured by mapping the finite-dimensional raw data to a high-dimensional feature space.In order to improve the performance of the proposed method,we introduce a local density of Gaussian kernel calculation data points.The low-dimensional to high-dimensional mapping is achieved through the Gaussian kernel distance,so GKDPCA can detect non-spherical clustering,and GKDPCA is more suitable for small data sets,and the selection of clustering centers is more accurate.(2)An improved conditional generation adversarial network method is proposed to solve the problem of data imbalance.In this paper,the clustering model is used in CGAN to avoid the noise of oversampling data.The corresponding samples are generated by inputting specific condition variables,which eliminates the imbalance between the distribution of social robots and the interior.At the same time,by introducing Wasserstein distance with gradient penalty,the CGAN convergence judgment conditions are improved,and the problems of model collapse and gradient disappearance in traditional CGAN are solved.(3)Design and implement a social robot detection system.Based on the proposed method,this article designs and implements a social robot detection system,which mainly implements data detection and account detection functions.The social robot detection system mainly includes a data collection module,a feature extraction module,a classifier model training module,and a social robot detection module.It also introduces the main functions and implementation processes of each module.Finally,in order to verify the effectiveness and stability of the system,the robot detection system was tested.The paper compares the improved CGAN with three common oversampling algorithms,and studies the effects of imbalance and the sampling ratio of the original data on oversampling.The experimental results show that the performance of the improved CGAN is better than other sampling methods.F1 score,G-mean,AUC received higher evaluation scores.At the same time,the test results for the social robot detection system show that the system can accurately identify the category of the account to be detected according to user input.The system improves the imbalance characteristics of the original data set through improved CGAN,and uses the balanced data set to train a stable classifier.The system identifies the input sample category of the user through a stable classifier.