Data Analytics for Harmful Online Communication in Social Media

Nowadays, our lives are not anymore only shaped by the real-world but also by the digital sphere. The rise of social media has fundamentally changed how we consume, perceive and share information, how we interact with people and how we pursue our right to freedom of expression. The ability to exchange personal opinions freely and, if necessary, anonymously is a pillar of democracy. Moreover, the option to interact with people from all over the world almost instantaneously broadens our horizons and allows us to view controversial topics from different perspectives. However, there is no such thing as a free lunch. With all these freedoms, there come significant challenges. Online conversations are increasingly penetrated by abusive language inciting hatred against groups or individuals, causing significant emotional stress. Moreover, coordinated manipulation of personal opinion through the targeted spread of misinformation and disinformation is frequently observed, especially in large political events like elections. The amplification of this harmful content has been significantly simplified through easy access to programmatic interfaces and the implied capability of automation on social media platforms. Building resilience against these threats is one of the more significant challenges of our current generation and can be achieved by several means. From the human perspective, a deep understanding of the medium and potential attack vectors is a necessary prerequisite. From the computer science perspective, clear problem definitions and adequate ground-truth data are required. In this publication-based thesis, I address these societal threats sparked by harmful online communication from a data analytics point of view. I elaborate on current challenges specific to the domain and what we as researchers can and should do to improve the current situation. This goes beyond improving algorithms and pushing accuracy scores. It goes beyond the blind collection of more training data. Sometimes it is necessary to re-evaluate how we approach a problem in the first place, leading to new perspectives and ultimately to new solutions.