Visual Analytics for Graph Representation Learning
Problem
Graph representation learning (also known as graph embedding) is a powerful technique for mapping a high-dimensional data structure, such as multivariate or heterogeneous graphs, into a lower-dimensional vector space. These embeddings are widely used in machine learning tasks, including node similarity, relationship identification, clustering, recommendation, and prediction.
While graph embeddings are effective for downstream tasks, the resulting representations are often opaque to human interpretation. In particular, it is usually unclear why two nodes are close (or far apart) in the embedding space, which graph structures or node attributes contribute to this similarity or dissimilarity, and how the embedding relates back to the original graph topology. This lack of interpretability poses both a challenge and an opportunity for visual analytics.
Aim
The goal of this project is to develop a visual analytics that helps users to understand, explore, and compare graph embeddings. The approach should support users in interpreting node and edge features, neighborhoods, clusters, and relative positions in the embedding space by linking them back to the structural and semantic properties of the original graph.
The student will gain experience with graph-structured data and graph embedding techniques, and will work with one (or more) graph datasets, including (but not limited to):
- Art History / Cultural Heritage: Exhibitions of Modern European Painting 1905-1915
- Molecules and Biomedical Networks: TUDatasets
The student is expected to complete the following three main tasks:
- Extracting and preparing graph data from an application domain (e.g., art history, biomedical networks, or knowledge graphs)
- Applying graph representation learning (embedding) techniques, using libraries such as PyTorch Geometric (PyG), Deep Graph Library (DGL), or PyKEEN
- Designing and implementing interactive visualizations to support the exploration, interpretation, and comparison of graph embeddings, bridging the gap between abstract embedding spaces and interpretable graph semantics.
Other information
Qualifications
- Strong interest in graph visualization, visual analytics, and graph-based (network) data
- Knowledge of (or willingness to learn) network analysis and machine learning, particularly graph embedding and representation learning
- Some background or experience in web application development, ideally using the React.js framework
- Familiarity with web-based visualization libraries such as Vega, Vega-Lite, or D3.js (or willingness to learn these tools); the ability to work with WebGL is a plus
References
- Hamilton, W. L. Graph Representation Learning. Morgan & Claypool Publishers, 2020. https://www.cs.mcgill.ca/~wlh/grl_book/files/GRL_Book.pdf
- Khoshraftar, S., & An, A. (2024). A Survey on Graph Representation Learning Methods. ACM Transactions on Intelligent Systems and Technology, 15(1), 1-55. https://doi.org/10.1145/3633518
Filipov, V., Arleo, A., & Miksch, S. (2023, September). Are We There Yet? A Roadmap Of Network Visualization From Surveys To Task Taxonomies. In Computer Graphics Forum (Vol. 42, No. 6, p. e14794). https://www.cvast.tuwien.ac.at/bibcite/reference/558
Huang, Z., Witschard, D., Kucher, K., & Kerren, A. (2023, June). VA+ Embeddings STAR: A State‐of‐the‐Art Report on the Use of Embeddings in Visual Analytics. In Computer Graphics Forum (Vol. 42, No. 3, pp. 539-571). https://doi.org/10.1111/cgf.14859
Wang, Q., Huang, K., Chandak, P., Zitnik, M., & Gehlenborg, N. (2022). Extending the Nested Model for User-Centric XAI: A Design Study on GNN-based Drug Repurposing. IEEE Transactions on Visualization and Computer Graphics, 29(1), 1266-1276. https://zitniklab.hms.harvard.edu/publications/papers/DrugExplorer-ieeevis22.pdf
Ren, D., Hohman, F., Lin, H., & Moritz, D. (2025). Embedding Atlas: Low-Friction, Interactive Embedding Visualization. arXiv preprint arXiv:2505.06386. https://arxiv.org/abs/2505.06386 https://github.com/apple/embedding-atlas