Subgraph-Informed Scoring Boosts Knowledge Graph Completion Accuracy

Global: Subgraph-Informed Scoring Boosts Knowledge Graph Completion Accuracy

Overview

A new framework named SLogic has been introduced to assign query‑dependent scores to logical rules used in knowledge graph completion. By evaluating the local subgraph surrounding a query’s head entity, SLogic determines the relevance of each rule for that specific instance, addressing the limitation of uniformly weighted rule schemas in earlier approaches.

Background

Logical rule‑based methods have long been valued for their interpretability, as they encode compositional relationships in human‑readable inference rules. However, most existing systems apply a single confidence weight to each rule schema across all queries, ignoring variations in rule importance that arise from differing graph contexts.

Method Overview

SLogic’s core component is a context‑aware scoring function that extracts a subgraph defined by the head entity of a given query. The function analyzes structural features of this subgraph to compute a score that reflects how pertinent a particular rule is to the current query. This dynamic weighting enables the system to prioritize rules that are more likely to contribute to accurate inference in the local neighborhood.

Experimental Evaluation

Extensive experiments on standard benchmark datasets demonstrate that SLogic consistently outperforms prior rule‑based techniques. In head‑to‑head comparisons, the framework also achieves performance that is competitive with leading neural and embedding‑based baselines, indicating that query‑dependent weighting can narrow the gap between interpretability and accuracy.

Interpretability

Beyond quantitative gains, SLogic generates query‑specific, human‑readable logical rules that serve as explicit explanations for each inference. This feature preserves the transparency of rule‑based models while providing users with insight into the reasoning process for individual predictions.

Implications and Future Work

The results suggest that incorporating local graph context into rule scoring can enhance both the effectiveness and explainability of knowledge graph completion systems. Researchers anticipate extending the approach to larger, heterogeneous graphs and exploring integration with other symbolic‑neural hybrid models.

This report is based on information from arXiv, licensed under Academic Preprint / Open Access. Based on the abstract of the research paper. Full text available via ArXiv.