Uncertainty in Representation Learning on Knowledge Graphs

2026-10-03

AI Summary

This study addresses the lack of reliability guarantees and the multifaceted uncertainties inherent in existing knowledge graph embedding (KGE) predictions. It systematically investigates epistemic, algorithmic, and predictive uncertainties within KGE frameworks and proposes a model-agnostic voting aggregation approach. Furthermore, this work introduces conformal prediction to the KGE domain for the first time, integrating statistical confidence interval estimation with embedding-based probabilistic reasoning to provide distribution-free theoretical coverage guarantees for predictions. By establishing uncertainty-aware knowledge graph reasoning with rigorous theoretical assurances, this research significantly enhances the trustworthiness and reliability of KGE-based predictions.

Abstract

Knowledge graph embedding (KGE) methods represent entities and predicates in continuous vector spaces to infer missing knowledge. Despite strong benchmark performance, their predictions often lack principled reliability guarantees, limiting their use in high-stakes applications. Moreover, uncertainty arises throughout the KGE pipeline, from incomplete or probabilistic input knowledge to stochastic training and prediction. This thesis systematically investigates three sources of uncertainty in KGE: knowledge uncertainty, arising from incomplete, noisy, or probabilistic input knowledge; algorithmic uncertainty, induced by randomness in model training; and predictive uncertainty, concerning the reliability of model outputs. To address algorithmic uncertainty, the thesis demonstrates that models trained under identical settings can produce substantially different predictions and introduces a voting-based aggregation framework to mitigate this instability. To quantify predictive uncertainty, it adapts conformal prediction to KGE, constructing answer sets with distribution-free coverage guarantees and extending them to provide predicate-conditional reliability guarantees. To support reasoning under knowledge uncertainty, it develops statistically valid prediction intervals for confidence-scored triples and an embedding-based approach to approximate probabilistic reasoning over statistical ontologies with formal soundness guarantees. Together, these complementary, model-agnostic methods provide a practical and theoretically grounded approach to uncertainty in KGE, advancing beyond predictive accuracy toward reliable and uncertainty-aware knowledge graph reasoning.

Problem

• Knowledge Graph Embedding
• Uncertainty Quantification
• Predictive Uncertainty
• Algorithmic Uncertainty
• Reliable Reasoning

Innovation

• Knowledge Graph Embedding
• Uncertainty Quantification
• Conformal Prediction
• Voting-based Aggregation
• Probabilistic Reasoning