Social Media Bot Detection

Sammanfattning: Social media platforms have revolutionized how people interact with each other and how people gain information. However, social media platforms such as Twitter and Facebook quickly became the platform for public manipulation and spreading or amplifying political or ideological misinformation. Although malicious content can be shared by individuals, today millions of individual and coordinated automated accounts exist, also called bots which share hate, spread misinformation and manipulate public opinion without any human intervention. To make things worse, the sophistication of these automated accounts is so high that they become unidentifiable by humans and difficult for algorithms to detect. In the past decade, researchers have developed novel methods to detect harmful, machinecontrolled accounts. Most of these methods are supervised and are based on classical machine learning approaches. Recently, unsupervised and adversarial methods have also been researched. What is usually common in all these methods is that the detection is based on account level meta-data and feature extraction. The work presented in this thesis, aims at designing, implementing and evaluating bot detectors that are based on deep-learning models and which do not require account level meta-data nor feature extraction. The thesis work relies on the PAN 2019 Bots and Gender Profiling task dataset where Twitter accounts had to be classified as humans or bots. Furthermore, the thesis shows that deep-learning models can yield an accuracy of 0.89 on the PAN 2019 Bots and Gender Profiling dataset; thus, they can compete with classical machine learning methods. Moreover, the findings of this work also show that pre-trained models will be able to improve the accuracy of deep-learning models.