Predicate-Conditional Conformalized Answer Sets for Knowledge Graph Embeddings

TL;DR

This paper introduces CondKGCP, a method that provides predicate-conditional uncertainty estimates for knowledge graph embeddings, ensuring more reliable and query-specific confidence in high-stakes applications.

Contribution

CondKGCP is a novel approach that approximates predicate-conditional coverage guarantees while maintaining compact answer sets in knowledge graph embeddings.

Findings

CondKGCP achieves predicate-conditional coverage guarantees.

Empirical results show CondKGCP produces reliable and compact answer sets.

Theoretical proofs support the method's coverage guarantees.

Abstract

Uncertainty quantification in Knowledge Graph Embedding (KGE) methods is crucial for ensuring the reliability of downstream applications. A recent work applies conformal prediction to KGE methods, providing uncertainty estimates by generating a set of answers that is guaranteed to include the true answer with a predefined confidence level. However, existing methods provide probabilistic guarantees averaged over a reference set of queries and answers (marginal coverage guarantee). In high-stakes applications such as medical diagnosis, a stronger guarantee is often required: the predicted sets must provide consistent coverage per query (conditional coverage guarantee). We propose CondKGCP, a novel method that approximates predicate-conditional coverage guarantees while maintaining compact prediction sets. CondKGCP merges predicates with similar vector representations and augments calibration with rank information. We prove the theoretical guarantees and demonstrate empirical effectiveness of CondKGCP by comprehensive evaluations.