Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Logical Reasoning with Relation Network for Inductive Knowledge Graph Completion

Qinggang Zhang, Keyu Duan, Junnan Dong, Pai Zheng, Xiao Huang

TL;DR

This work addresses inductive knowledge graph completion under evolving KGs by introducing NORAN, a relation-network–driven framework that centers on relation semantics to infer missing relations for unseen entities. NORAN constructs a relation network where each triple is a node linked by shared entities, and employs relational message passing on both the ego-graphs and the relation network, guided by a mutual-information–based objective (LEIM) to preserve logic-like evidence. The approach is model-agnostic with theoretical guidance on when to use learnable versus fixed MP layers, and is validated on five real-world KG benchmarks where it achieves substantial gains over state-of-the-art baselines. The results highlight the practical impact of modeling entity-independent relational evidence and translating logical patterns into inductive KG reasoning, enabling more robust handling of unseen entities in dynamic knowledge graphs.

Abstract

Inductive knowledge graph completion (KGC) aims to infer the missing relation for a set of newly-coming entities that never appeared in the training set. Such a setting is more in line with reality, as real-world KGs are constantly evolving and introducing new knowledge. Recent studies have shown promising results using message passing over subgraphs to embed newly-coming entities for inductive KGC. However, the inductive capability of these methods is usually limited by two key issues. (i) KGC always suffers from data sparsity, and the situation is even exacerbated in inductive KGC where new entities often have few or no connections to the original KG. (ii) Cold-start problem. It is over coarse-grained for accurate KG reasoning to generate representations for new entities by gathering the local information from few neighbors. To this end, we propose a novel iNfOmax RelAtion Network, namely NORAN, for inductive KG completion. It aims to mine latent relation patterns for inductive KG completion. Specifically, by centering on relations, NORAN provides a hyper view towards KG modeling, where the correlations between relations can be naturally captured as entity-independent logical evidence to conduct inductive KGC. Extensive experiment results on five benchmarks show that our framework substantially outperforms the state-of-the-art KGC methods.

Logical Reasoning with Relation Network for Inductive Knowledge Graph Completion

TL;DR

This work addresses inductive knowledge graph completion under evolving KGs by introducing NORAN, a relation-network–driven framework that centers on relation semantics to infer missing relations for unseen entities. NORAN constructs a relation network where each triple is a node linked by shared entities, and employs relational message passing on both the ego-graphs and the relation network, guided by a mutual-information–based objective (LEIM) to preserve logic-like evidence. The approach is model-agnostic with theoretical guidance on when to use learnable versus fixed MP layers, and is validated on five real-world KG benchmarks where it achieves substantial gains over state-of-the-art baselines. The results highlight the practical impact of modeling entity-independent relational evidence and translating logical patterns into inductive KG reasoning, enabling more robust handling of unseen entities in dynamic knowledge graphs.

Abstract

Inductive knowledge graph completion (KGC) aims to infer the missing relation for a set of newly-coming entities that never appeared in the training set. Such a setting is more in line with reality, as real-world KGs are constantly evolving and introducing new knowledge. Recent studies have shown promising results using message passing over subgraphs to embed newly-coming entities for inductive KGC. However, the inductive capability of these methods is usually limited by two key issues. (i) KGC always suffers from data sparsity, and the situation is even exacerbated in inductive KGC where new entities often have few or no connections to the original KG. (ii) Cold-start problem. It is over coarse-grained for accurate KG reasoning to generate representations for new entities by gathering the local information from few neighbors. To this end, we propose a novel iNfOmax RelAtion Network, namely NORAN, for inductive KG completion. It aims to mine latent relation patterns for inductive KG completion. Specifically, by centering on relations, NORAN provides a hyper view towards KG modeling, where the correlations between relations can be naturally captured as entity-independent logical evidence to conduct inductive KGC. Extensive experiment results on five benchmarks show that our framework substantially outperforms the state-of-the-art KGC methods.
Paper Structure (30 sections, 1 theorem, 8 equations, 1 figure, 5 tables, 1 algorithm)

This paper contains 30 sections, 1 theorem, 8 equations, 1 figure, 5 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Transductive KG Completion
  4. Inductive KG Completion
  5. Problem Formulation
  6. Methodology
  7. Relation Network Construction
  8. Relation network
  9. Interpretation with logic reasoning
  10. Relational Message Passing
  11. Feature initialization
  12. Message passing
  13. Mutual information maximization
  14. Training Scheme and KG Inference
  15. Training objective
  16. ...and 15 more sections

Key Result

Corollary 1

Given a graph $\mathcal{G}=(\mathcal{V}, \mathcal{E})$, let $\bm{h}_v^{(i)}$ be the hidden features of node $v$ at $i$-th layer of a $k$-layer GNN model $\Psi$ whose MP can be rewritten as $\bm{X}^{(i+1)} = \sigma( \mathcal{C}^{(i)}\bm{X}^{(i)}\circ f^{(k)} )$. If the convolution matrix is pre-fixed

Figures (1)

  • Figure 1: (i) A toy example of inductive knowledge graph completion, i.e. predicting the relationship ("?") for unseen entities, e.g. "Martin Eberhard", provided with a few links, e.g. (Martin, :WorkAt, TESLA); (ii) Illustration of the corresponding relation network for knowledge graph, which regards each triple as a relational node and thus could aggregate context information for inductive inference without expensive retraining look-up embedding tables as embedding-based paradigm.

Theorems & Definitions (6)

  • Definition 1
  • Definition 2
  • Definition 3
  • Corollary 1
  • Remark 1
  • Remark 2