An Empirical Study of Machine learning Algorithms for Social Media Bot Detection

Abstract

Social media bots can change society's perspective in different aspects of life. This paper analyzes sentiment features and their effect on the accuracy of machine learning models for social media bot detection. Social bots can use tweet sentiment to create a backfire effect and confirmation bias to create a fake trend or change public opinion. We analyze bot detection problems based on sentiment features inspired by the work by Micheal Workman [1] and create new features based on textual information of online comments. We offer a quantitative approach to create new features and compare machine learning models for bot detection. This work is based on psychological and social effects inherent in tweets' text content based on the work by [1]. The new set of sentiment features are extracted from a tweet's text and used to train bot detection models. Also, we implement the new model for the Dutch language and achieve more than 87% accuracy for the Dutch tweets based on new sentiment features. Considering new sentiment features based on psychological and social factors for a tweet's text will open a potential research area for social media bot detection.