Tuning machine learning models to detect bots on Twitter

Abstract

Bot generated content on social media can spread fake news and hate speech, manipulate public opinion and influence the community on relevant topics, such as elections. Thus, bot detection in social media platforms plays an important role for the health of the platforms and for the well-being of societies. In this work, we approach the detection of bots on Twitter as a binary output problem through the analysis of account features. We propose a pipeline for feature engineering and model training, tuning and selection. We test our pipeline using 3 publicly available bot datasets, comparing the accuracy of all trained models with the model selected at the end of our pipeline.