Detecting Social Bots on Twitter with Improved Conditional Generative Adversarial Networks

PROJECT TITLE :

Using Improved Conditional Generative Adversarial Networks to Detect Social Bots on Twitter

ABSTRACT:

The detection and elimination of dangerous social bots in social media has piqued commercial and academic interest. The commonly used Machine Learning-based bot detection method results in an imbalance in the amount of samples in different categories. Minority samples are detected at a low rate due to classifier bias. To increase the detection accuracy of social bots, we propose an improved conditional generative adversarial network (improved CGAN) to extend imbalanced data sets before applying training classifiers. We propose a modified clustering approach, the Gaussian kernel density peak clustering algorithm (GKDPCA), to construct an auxiliary condition. This algorithm avoids the development of data-augmentation noise and reduces imbalances between and among social bot class distributions. We also introduce the Wasserstein distance with a gradient penalty to improve the CGAN convergence judgment condition, which solves model collapse and gradient disappearance in the classic CGAN. In experiments, three typical oversampling techniques are compared. The impacts of the original data's imbalance degree and expansion ratio on oversampling are investigated, and the improved CGAN outperforms the others. Experimental results reveal that the modified CGAN achieves higher evaluation scores in terms of F1-score, G-mean, and AUC when compared to three typical oversampling techniques.

Did you like this research project?

To get this research project Guidelines, Training and Code... Click Here