eXpath: Explaining Knowledge Graph Link Prediction with Ontological Closed Path Rules

TL;DR

EXpath is introduced which incorporates the concept of relation path with ontological closed path rules to enhance both the efficiency and effectiveness of LP interpretation and can be fused with other single-link explanation approaches to achieve a better overall solution.

Abstract

Link prediction (LP) is crucial for Knowledge Graphs (KG) completion but commonly suffers from interpretability issues. While several methods have been proposed to explain embedding-based LP models, they are generally limited to local explanations on KG and are deficient in providing human interpretable semantics. Based on real-world observations of the characteristics of KGs from multiple domains, we propose to explain LP models in KG with path-based explanations. An integrated framework, namely eXpath, is introduced which incorporates the concept of relation path with ontological closed path rules to enhance both the efficiency and effectiveness of LP interpretation. Notably, the eXpath explanations can be fused with other single-link explanation approaches to achieve a better overall solution. Extensive experiments across benchmark datasets and LP models demonstrate that introducing eXpath can boost the quality of resulting explanations by about 20% on two key metrics and reduce the required explanation time by 61.4%, in comparison to the best existing method. Case studies further highlight eXpath's ability to provide more semantically meaningful explanations through path-based evidence.

Figures (5)

• Figure 1: An example of material KG for synthesis route inference. To explain the predicted link ⟨material: BEMHUX, hasSolvent, solvent: dichloromethane⟩ (the dotted red link on the top), two key KG paths (blue links on the middle/bottom) are detected by our method: BEMHUX and dichloromethane sharing the same material sub-structure; BEMHUX appearing in the same paper with another material BEMHIL, which also uses dichloromethane as the solvent. Classical LP explanations (e.g., Kelpie) will select the single-hop links as explanations (thickened blue links).
• Figure 2: Pipeline of eXpath. (a) Path Aggregation: Identifies paths between $h$ and $t$ using breadth-first search (BFS) and compresses them into relation paths. (b) Path-based Rule Mining: Prunes relevant relation paths and selects high-confidence closed path (CP) and property transition (PT) rules. (c) Critical Fact Selection: Scores candidate facts based on rule relevance and confidence, selecting the highest-scoring facts for the final explanation.
• Figure 3: Principles and instances of ontological rules used in our framework. closed path (CP) rules describe the relationship between entities $X$ and $Y$ through alternative paths, while Property Transition (PT) rules capture transitions between different attributes of the same entity. These ontological rules are not predefined but are generalized patterns mined from the knowledge graph, supported by substructures that conform to the specified patterns.
• Figure 4: Average times in seconds to extract an explanation for Kelpie and eXpath.
• Figure 5: Explanation of the fact $\langle$$e_3$, actor, Jonathan_Pryce$\rangle$ predicted by LP models (ComplEx); (a) all 3-hop paths from head entity to tail entity. (b) Twelve high-confidence rules with $SC \geq 0.1$ identified by eXpath; (c) comparison of the explanation provided by KGEAttack (in purple edge), Kelpie (in green edges), and eXpath (in yellow edges).