Language-Agnostic Twitter-Bot Detection

Proceedings of Recent Advances in Natural Language Processing, pages 550–558, Varna, Bulgaria, Sep 2–4, 2019. https://doi.org/10.26615/978-954-452-056-4_065 550 Language-Agnostic Twitter Bot Detection J¨urgen Knauth Institute for Computer Science University of G¨ottingen jknauth@uni-goettingen.de Abstract In this paper we address the problem of de- tecting Twitter bots. We analyze a dataset of 8385 Twitter accounts and their tweets consisting of both humans and different kinds of bots. We use this data to train machine learning classifiers that distin- guish between real and bot accounts. We identify features that are easy to extract while still providing good results. We an- alyze different feature groups based on ac- count specific, tweet specific and behav- ioral specific features and measure their performance compared to other state of the art bot detection methods. For easy future portability of our work we focus on language-agnostic features. With Ad- aBoost, the best performing classifier, we achieve an accuracy of 0.988 and an AUC of 0.995. As the creation of good train- ing data in machine learning is often dif- ficult - especially in the domain of Twit- ter bot detection - we additionally analyze to what extent smaller amounts of train- ing data lead to useful results by review- ing cross-validated learning curves. Our results indicate that using few but expres- sive features already has a good practical benefit for bot detection, especially if only a small amount of training data is avail- able. 1 Introduction While at the end of the 90s digital communica- tion was still handled exclusively via well-defined Internet protocols, a completely new form of com- munication has established over the decades us- ing web-based technology: Social media. These are web-based platforms that allow users to pub- lish and exchange messages with each other after registration. These platforms are highly attractive: on the one hand, they establish a form of social community that enables people to enter into new relationships with each other or to maintain exist- ing relationships. On the other hand, these plat- forms are designed in such a way that their users can easily find new content based on algorithmic recommendations. Today, these social media platforms cover a wide variety of media. This work focuses on Twit- ter1 - a service that enables the publication of short text comments. As information and opinions are widely dissem- inated and exchanged via these platforms, the rise of social media as mass communication tools has increasingly attracted the interest of different ac- tors who try to use social media as influencing fac- tors. Businesses recognize the potential to adver- tise their products in this way at a very early stage, while political actors try to promote their views. With the rise of such platforms to mass media and the desire to influence public perception and opinion, a comparatively new phenomenon has de- veloped: The use of bots. In addition to normal human users, some social media platforms also feature computer programs that generate and pro- vide content. This phenomenon has been found on Twitter as …

Content truncated to 500 words