Master’s Thesis Opportunity: How Does TikTok Shape What Young Adults See About Health?

29.04.2026, Studentische Hilfskräfte, Praktikantenstellen, Studienarbeiten

More and more young adults in Germany are turning to TikTok for information on topics such as mental health, nutrition, fitness, and medical advice. But how does the platform's recommendation algorithm decide what health-related content users encounter, and what does that mean for their experience? Does TikTok's algorithm nudge users toward particular kinds of content over time? Do users notice when their feed shifts, and how does what they see relate to how they feel?

These questions connect public health, digital ethics, and computational social science, and they are difficult to answer with conventional research designs. TikTok's short-form, multimodal environment and opaque recommendation system mean that standard surveys or content audits capture only part of what is happening.

HARMONY – Our Project

HARMONY combines four complementary data layers: behavioral trace data from participants' TikTok exports, a large corpus of videos scraped and transcribed from their feeds, experience sampling of in-the-moment reactions, and a detailed baseline survey on attitudes, media use, and well-being. Our design links these streams around a shared timeline, connecting what users were shown, what that content contained, how they felt afterward, and who they are as individuals.

Two features of the design are worth noting. First, all participants use freshly created TikTok accounts, so the study captures how the algorithm learns and adapts from the very first interaction. Second, TikTok's data export logs every video the algorithm served, not only those the user tapped on, meaning the data reflect the feed as it was assembled, not merely what the user chose to acknowledge.

Thesis Directions

Direction A: Making the Content Analyzable
The raw data, videos, transcripts, and metadata need to become a structured, research-ready corpus. This means cleaning and organizing transcripts at scale, developing methods to identify what a video is about, and assessing how large language models can assist with systematic annotation. The goal is a content infrastructure that lets us and future researchers ask substantive questions of the data.

Direction B: Connecting Content to Experience
Once we know what users are seeing, the next question is what difference it makes. This direction involves modeling how content exposure changes over time, identifying moments where the algorithm shifts what it serves, and linking content patterns to participants' self-reported affect and sense of control. The aim is to move from describing feed composition toward understanding its experiential consequences.

Direction C: Who Sees What, and Why?
Before the observation period, all participants completed a survey covering political attitudes, health beliefs, trust in information sources, algorithmic literacy, media use, and well-being. This direction asks a different set of questions: which individual characteristics predict exposure to particular content types? Do users who are more algorithmically aware see different things? How do personality and attitudinal factors relate to what the recommendation system serves over time? This direction suits students who prefer working with survey instruments and statistical models rather than computational pipelines.

Students may also propose additional directions, provided they engage at least one of the project's data layers and can be operationalized within the HARMONY empirical framework.

What We Offer

Access to a largely collected dataset combining four linked data layers: behavioral logs, video content, experience sampling, and survey measures, including validated scales on attitudes, algorithmic literacy, and well-being. An interdisciplinary team spanning digital bioethics, media psychology, and computational social science, with supervision adapted to your chosen direction. Where appropriate, there is the possibility of contributing to publications or conference presentations. We cannot provide hands-on technical or statistical support (debugging, engineering assistance, data analysis); students should enter the thesis with the skills needed to implement their chosen direction independently.

Preferred Background

Directions A and B: Strong Python skills; experience with NLP, text classification, or representation learning (e.g., word embeddings, transformer-based embeddings, topic modeling); comfort working with large, heterogeneous multimodal datasets; familiarity with data pipeline design and reproducible workflows; ability to work independently on methodological problems.

Direction C: Background in sociology, psychology, political sciences, social sciences, statistics, or related fields; experience with psychometric methods (e.g., exploratory and confirmatory factor analysis, scale validation); familiarity with regression, multilevel models, or structural equation modeling; interest in individual differences, attitudes, and digital media use.

Apply

Please send your CV, Transcript of Records, and a short intro (under 400 words) describing your background, motivation, and interested direction to Valérie Nowak (valerie.nowak@tum.de) and Alexander Sobieska (alexander.sobieska@tum.de).

Kontakt: alexander.sobieska@tum.de