Emerging Online Phenomena • Online Harms • Auditing Sociotechnical Systems

We are looking for motivated BSc and MSc thesis students to work with us on research on Computational Social Science and at the intersection of Computational Social Science and Security. The focus is on emerging online phenomena, online harms, and methods to measure, audit, and understand sociotechnical systems such as online communities, platforms and AI systems.

P.S. You do not need to have a fully formed idea. Motivation and willingness to learn are the most important requirements.

What are 'Online Harms'?

Online harms broadly refer to negative societal, psychological, or security-related outcomes that arise from online platforms, communities, and AI systems. At INET, we lead studies on harms originating from communities, platform algorithms, and AI-driven systems.

A) Community-driven Harms & Online subcultures

Are you curious about how online communities shape behavior, norms, and harmful outcomes?

Example questions:
 - How do harmful narratives evolve in fringe or high-risk communities (e.g., Reddit, Online forums)?
 - What role does language play in sustaining or amplifying harmful norms?
 - How can we demystify and better understand the role of content moderation?

Example work from our group:
From Isolation to Desolation: Investigating Self-Harm Discussions in Incel Communities

B) Harms from Recommender Systems & Personalization of Apps

Are you curious about what social media platforms actually recommend to users and whether some demographics are disproportionately exposed to harmful content?

Typical work includes:
 - Auditing recommendation systems using sockpuppets or automated agents
 - Building scalable auditing infrastructure and feed collection
 - Measuring content moderation enforcement in practice

Example focus areas include improving existing auditing frameworks developed in our group that audit Instagram, TikTok, and other social media platforms.

C) LLMs: Bias, Censorship, and Real-world Use

Are you curious about how large language models behave across languages, regions, and domains?

Example questions:
 - Does LLM censorship differ across languages or geographic vantage points?
 - When do safety policies lead to information access disparities?
 - How do people actually use LLMs in sensitive domains?

Example work: Our group has work to appear in NDSS 2026 on LLM content moderation across languages and vantage points (reach out to know more).

D) Harms from Generative AI systems

Are you curious about how generative AI changes the scale and nature of online harm?

Possible topics:

- GenAI search inconsistencies (see example work below)
 - Measurement and detection of AI-generated harmful artifacts
     - AI-generated scams and social engineering material
     - Non-consensual synthetic intimate imagery

Example work:
 Auditing Google's AI Overviews and Featured Snippets: A Case Study on Baby Care and Pregnancy. (to appear in ICWSM’26)

E) Computational Social Science Measurement Studies

Are you interested in “traditional” computational social science research focused on careful measurement?

Example topics:
 - Community lifecycle analysis on Reddit
 - Mental health support forums and peer interactions
 - Cross-platform information diffusion like information spilling from one platform to another

Example works:

• Auditing radicalization pathways on YouTube | Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency
• Effects of Research Paper Promotion via ArXiv and X | Proceedings of the International AAAI Conference on Web and Social Media
• A Sticker is Worth a Thousand Words: Characterizing the Use and Abuse of Stickers on WhatsApp Political Groups in Brazil   

F) A topic of your own choosing

If you have observed an interesting online phenomenon, a sociotechnical challenge, or something that personally fascinates you, you are encouraged to propose it. Motivation and curiosity matter more than a polished idea.

Places to find and (maybe) refine thesis ideas

Students are encouraged to explore recent work from top venues such as:
 - ICWSM, WWW / The Web Conference, CSCW, CHI 
 - USENIX Security, CCS, NDSS, IEEE S&P (social and socio-technical tracks)

 A strong thesis typically combines: a clear phenomenon, a measurable question, and an executable methodology (given the time constraints, ofc).

Who should apply?

If you are interested in online platforms, communities, AI, and/or security topics, and are willing to engage with data, methods, and research papers. Prior experience with Python, NLP, network analysis and quantitative measurement is helpful but not required.

Contact

If you are interested or want to discuss stuff, please reach out with:
 - Your degree program (BSc/MSc) and timeline
 - A short description of which topic(s) interest you
 - Relevant skills or background

Coordinator(s): Moonis Ali, Breno de Sousa Matos
For queries, reach out to: moonis.ali@mpi-inf.mpg.de