Performance Comparison of Deep Learning Algorithms in Twitter Bot Detection

Difficulty in detecting social media bots - software-controlled accounts that seem as human - has serious consequences. Bots have been used to affect the financial market, political elections, and spreading conspiracy theories. Most approaches detect bots at the account level by studying social media posts and using network structure, system dynamics, content analytics, etc. In this study, we compare the performance of the recurrent neural network (RNN), convolutional neural network (CNN), and feed forward neural network (FNN) in detecting Twitter bots by evaluating the textual content of 10,416,240 tweets from humans and bots. Deep learning models utilize sample data to gauge performance. The CNN model identified recent social bot samples most accurately (97.67 %). In comparison, the RNN model achieved great accuracy (97.04 %) with a longer training time. RNN and CNN algorithms perform well when compared to other deep learning classifiers.