Motivation 

Ensuring diversity in media offerings is central to ZDF’s mission. This project explores how embedding techniques can quantify and visualize content diversity, supporting data-driven decisions for strategic portfolio planning. 

Goals
This TUM-DI-LAB project aims to generate embeddings for ZDF content using language or multimodal models, evaluate diversity across the portfolio, and identify gaps informed by usage data. This analysis will guide editorial decisions to maintain variety and fairness. 

As part of the project team, you will: 

1. Design and implement embeddings using state-of-the-art models (e.g., BERT, multimodal architectures).
2. Apply clustering and dimensionality reduction to create interpretable visualizations of thematic coverage based on metadata and learned representations.
3. Integrate and analyze usage data to derive actionable recommendations.

Comparisons between metadata-based and learned embeddings will highlight trade-offs in explainability and performance.

Requirements  

We are looking for students with the following skills and interests: 

• Analytical Thinkers with a Media Mindset: Ability to deconstruct complex problems, breaking them down into actionable steps, and crafting innovative solutions. An interest in media and its societal impact will set you apart.
• Programming: Strong proficiency in Python and experience with machine learning or data visualization. Confidence working with large datasets is a plus.
• Curiosity & Drive: You’re passionate about learning, exploring new ideas, and pushing boundaries. You want your work to matter - helping shape diverse and engaging media for millions.

Join us to shape the future of public-service media and contribute to a more informed world. Apply now! 

Apply to this project here