Large Language Models (LLMs) have revolutionized natural language processing, yet they struggle to interpret and reason over the highly structured, interconnected data that powers modern enterprises. Knowledge Graphs (KGs) are a powerful mechanism for representing this structured knowledge, but building and utilizing them at scale remains a challenge. A new frontier is emerging with Graph Foundation Models (GFMs), which promise to bridge the gap between the generative power of LLMs and the structured reasoning of KGs.
The vision of this thesis is to leverage the application of GFMs within the Bosch ecosystem. We aim to explore how these cutting-edge models can automate the construction, completion, and reasoning over our enterprise knowledge graphs.
-
During your thesis you will conduct a comprehensive literature review of the state-of-the-art in Graph Foundation Models and their application. You will analyze existing benchmarks and datasets for knowledge graph construction, link prediction, and advanced graph-based analytics to identify key methodologies.
-
Furthermore, you will develop innovative models and experiment with their implementation. You will use GFMs to extract structured entities and their relationships from internal Bosch documents and fine-tune or prompt GFMs to infer and predict missing links and relationships within our existing knowledge graphs.
-
You will develop methods to translate natural language questions into formal graph queries or use the GFM to reason over graph pathways, directly supporting use cases like root-cause analysis in manufacturing.
-
Finally, you will rigorously evaluate the performance of the developed models on both standard academic benchmarks and on real-world Bosch datasets and use cases. You will analyze the scalability, robustness, and deployment potential of the developed methods within Bosch's enterprise environment.