Social Bot Detection based on Language-independent Stylometric Analysis using dual-channel multi-dimensional CNN

Nowadays, people use social media to interact with one another regularly. Social media platforms have received a great deal of criticism for their ability to democratize digital communication. Also, Social networks are often abused to bias public opinion and promote falsehoods. Unfortunately, there have been instances when misinformation attempts have been managed by bots, Social media accounts handled by computer scripts that seem to be real humans. Using user activity patterns, we propose an innovative approach for detecting social bots, which does not rely on communicative language. The efficacy of our approach is demonstrated through its application to two languages that have been extensively studied by many scholars in the past. To evaluate its performance, we employ the CLEF 2019 dataset, which consists of sub-datasets in both English and Spanish. This study employs a dual-channel multi-dimensional convolutional neural network (DCMDCNN) model for the identification of social bots. The model incorporates four stylometric features that primarily capture user interactions in a language-independent way. Our proposed model outperforms many of the techniques used in this goal, with the highest accuracy of 93.45%, 90.72%, and 92.34% for English, Spanish, and Aggregated versions respectively.