Predicting from a Different Perspective: A Re-ranking Model for Inductive Knowledge Graph Completion

TL;DR

The paper tackles inductive knowledge graph completion by showing that rule-length variation in path-based models causes divergent predictions. It introduces ReDistLP, a two-stage cascade using BERTRL variants trained on different hop-depth rules to re-rank candidate predictions, guided by fuzzy-set theory to maximize re-ranking effectiveness. Empirical results across three inductive benchmarks demonstrate improved accuracy on two datasets, highlighting the value of prediction diversity between retriever and re-ranker. The work also reveals that optimal rank-thresholding is crucial and suggests future work to refine threshold strategies for further gains. Overall, the approach offers a computationally efficient yet accurate mechanism for inductive KGC by combining diverse rule-based signals in a principled re-ranking framework.

Abstract

Rule-induction models have demonstrated great power in the inductive setting of knowledge graph completion. In this setting, the models are tested on a knowledge graph entirely composed of unseen entities. These models learn relation patterns as rules by utilizing subgraphs. Providing the same inputs with different rules leads to differences in the model's predictions. In this paper, we focus on the behavior of such models. We propose a re-ranking-based model called ReDistLP (Re-ranking with a Distinct Model for Link Prediction). This model enhances the effectiveness of re-ranking by leveraging the difference in the predictions between the initial retriever and the re-ranker. ReDistLP outperforms the state-of-the-art methods in 2 out of 3 benchmarks.

Figures (2)

• Figure 1: Visualization of predicted top-10 entities. Red nodes represent entities from the input triple. Green nodes represent predicted entities. Blue nodes represent entities that appear in connected paths.
• Figure 2: An overview of ReDistLP.