Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Inductive Knowledge Graph Completion with GNNs and Rules: An Analysis

Akash Anil, Víctor Gutiérrez-Basulto, Yazmín Ibañéz-García, Steven Schockaert

TL;DR

This work addresses inductive knowledge graph completion by dissecting why rule-based methods underperform GNNs and proposing hybrid strategies that combine rule-based evidence with graph neural networks. It introduces rule-instantiation graphs to summarize multiple ground rules and uses GNNs to aggregate their evidence, achieving performance close to state-of-the-art GNNs like NBFNet while preserving interpretability. Additionally, a variant that uses the full knowledge graph for reranking can surpass NBFNet, highlighting the potential to mitigate spurious correlations in purely neural approaches. The study also underscores methodological differences in evaluating inductive KG methods, calling for standardized benchmarks and reporting.

Abstract

The task of inductive knowledge graph completion requires models to learn inference patterns from a training graph, which can then be used to make predictions on a disjoint test graph. Rule-based methods seem like a natural fit for this task, but in practice they significantly underperform state-of-the-art methods based on Graph Neural Networks (GNNs), such as NBFNet. We hypothesise that the underperformance of rule-based methods is due to two factors: (i) implausible entities are not ranked at all and (ii) only the most informative path is taken into account when determining the confidence in a given link prediction answer. To analyse the impact of these factors, we study a number of variants of a rule-based approach, which are specifically aimed at addressing the aforementioned issues. We find that the resulting models can achieve a performance which is close to that of NBFNet. Crucially, the considered variants only use a small fraction of the evidence that NBFNet relies on, which means that they largely keep the interpretability advantage of rule-based methods. Moreover, we show that a further variant, which does look at the full KG, consistently outperforms NBFNet.

Inductive Knowledge Graph Completion with GNNs and Rules: An Analysis

TL;DR

This work addresses inductive knowledge graph completion by dissecting why rule-based methods underperform GNNs and proposing hybrid strategies that combine rule-based evidence with graph neural networks. It introduces rule-instantiation graphs to summarize multiple ground rules and uses GNNs to aggregate their evidence, achieving performance close to state-of-the-art GNNs like NBFNet while preserving interpretability. Additionally, a variant that uses the full knowledge graph for reranking can surpass NBFNet, highlighting the potential to mitigate spurious correlations in purely neural approaches. The study also underscores methodological differences in evaluating inductive KG methods, calling for standardized benchmarks and reporting.

Abstract

The task of inductive knowledge graph completion requires models to learn inference patterns from a training graph, which can then be used to make predictions on a disjoint test graph. Rule-based methods seem like a natural fit for this task, but in practice they significantly underperform state-of-the-art methods based on Graph Neural Networks (GNNs), such as NBFNet. We hypothesise that the underperformance of rule-based methods is due to two factors: (i) implausible entities are not ranked at all and (ii) only the most informative path is taken into account when determining the confidence in a given link prediction answer. To analyse the impact of these factors, we study a number of variants of a rule-based approach, which are specifically aimed at addressing the aforementioned issues. We find that the resulting models can achieve a performance which is close to that of NBFNet. Crucially, the considered variants only use a small fraction of the evidence that NBFNet relies on, which means that they largely keep the interpretability advantage of rule-based methods. Moreover, we show that a further variant, which does look at the full KG, consistently outperforms NBFNet.
Paper Structure (13 sections, 2 equations, 5 figures, 7 tables)

This paper contains 13 sections, 2 equations, 5 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Background
  4. Hybrid Link Prediction Strategies
  5. Reranking with GNNs
  6. Reranking with NBFNet
  7. Experiments
  8. Results
  9. Analysis
  10. Conclusions
  11. Acknowledgements
  12. Bibliographical References
  13. Appendix

Figures (5)

  • Figure 1: Considered hybrid strategies.
  • Figure 2: Hits@3 performance for frequent (10% most frequent), common (10-50% most frequent) and rare (50% least frequent) relations.
  • Figure 3: NELL-995 v2 (guide, organization dissolved at date, may).
  • Figure 4: WN18RR v1 (unsimilarity, derivationally related form, unlike).
  • Figure 5: NELL-995 v3 (wild animals, inv. animal is type of animal, tigers).