Knowledge Graph Foundation Models
SEMMA [1] introduces a dual-module foundation model that combines textual semantics with structure. By leveraging Large Language Models (LLMs) to generate semantic embeddings for relation identifiers, SEMMA constructs a "textual relation graph" that is fused with the traditional structural component.
This approach pays off in difficult scenarios: while structural methods often collapse when facing entirely new relation vocabularies, SEMMA utilizes the semantic meaning of these new relations to maintain high performance. Experiments across 54 diverse knowledge graphs show it is up to 2x more effective in these challenging generalization settings than purely structural baselines.
Uncertainty Quantification
In high-stakes applications, such as medical diagnosis or financial forecasting, knowing what a model predicts is often less important than knowing how confident it is in that prediction. Standard Uncertain Knowledge Graph Embedding (UnKGE) methods fail here: they output single-point estimates without quantifying the reliability of that score.
To address this, UnKGCP [2] proposes a framework that shifts from point estimates to prediction intervals with statistical guarantees. Building on the principles of conformal prediction, the method generates a range of values guaranteed to contain the true score with a user-specified level of confidence (e.g., 95%).
References
[1] Arvindh Arun, Sumit Kumar, Mojtaba Nayyeri, Bo Xiong, Ponnurangam Kumaraguru, Antonio Vergari, and Steffen Staab. 2025. SEMMA: A Semantic Aware Knowledge Graph Foundation Model. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 31813–31836, Suzhou, China. Association for Computational Linguistics.
[2] Yuqicheng Zhu, Jingcheng Wu, Yizhen Wang, Hongkuan Zhou, Jiaoyan Chen, Evgeny Kharlamov, and Steffen Staab. 2025. Certainty in Uncertainty: Reasoning over Uncertain Knowledge Graphs with Statistical Guarantees. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 8741–8763, Suzhou, China. Association for Computational Linguistics.
