About the Seminar
About the Course: Computational Analysis of Visual Social Media (Winter 2024/25)
This project-centered research seminar focuses on group work, lectures, and practical sessions. The seminar introduces students to various research methodologies, including data collection, content analysis, and machine learning techniques. Students will work with tools such as Whisper for video transcription, GPT, fine-tuning BERT variants, and multimodal models like GPT-4o. Groups will explore diverse research topics and datasets, such as political communication on social media, influencer content, and misinformation, guided by an interdisciplinary approach that integrates media studies, political science, and communication science. The seminar is designed for master students with basic programming experience.
Taught by Michael Achmann-Denkler, a PhD candidate at the University of Regensburg, pursuing a PhD titled “Decoding Digital Campaigns: Computational Insights into Multimodal Political Communication on Social Media.” This seminar is closely linked to his research, aiming to test approaches, methods, and ideas on a broader scale
Course Outcomes
Theoretical Skills: - Accessibility of social media data for researchers - Concepts of text and images as data - Ethical and legal limitations - Understanding social media as trace data
Practical Skills: - Collecting Instagram and TikTok content - Applying OCR and video transcription - Exploring and classifying text and images computationally - Evaluating and optimizing classifications using human annotations - Presenting research findings effectively
Requirements
- Engage independently with literature, formulate research questions, and develop operationalizations
- Willingness to master new tools via practical exercises and provided Jupyter Notebooks
- Active and regular collaboration within your project team
- Continuous documentation of progress through a project wiki
- Submit a project report by the end of the semester
Project Documentation
- Documentation will follow an Open Science approach
- Use the project report template for your final report
- Reports will be published on social-media-lab.net
Project Report
- Submit a collaborative report (approx. 20 pages)
- Follow the IMRAD structure: Introduction, Method, Results, and Discussion
- Use APA citation style
- Deadline: 31.03.2025
Project Ideas
- Video Analysis: Apply existing analysis approaches to video content
- Replication Study: Replicate Peng (2021) on a German/European dataset
- Scalability Study: Test how well methods tested on small datasets scale with large data
Materials & Tools
- Course materials are available on social-media-lab.net.
- Collaborative editing through GitHub using Quarto and Markdown
- All content will be under GPL-3
Schedule Highlights
- October: Introduction & Tool Setup (Colab, GitHub)
- November: Data Collection (Instagram, TikTok)
- December: Text Analysis & Operationalization, Text Classification using BERT
- January: Evaluation & Image Analysis, Visual Data Exploration with LLMs and Clustering
- February: Data Analysis as a Conversation, Visual Presentation of Results (RAWGraphs, Figma)