Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Context-aware Inductive Knowledge Graph Completion with Latent Type Constraints and Subgraph Reasoning

Muzhi Li, Cehao Yang, Chengjin Xu, Zixing Song, Xuhui Jiang, Jian Guo, Ho-fung Leung, Irwin King

TL;DR

The paper tackles inductive knowledge graph completion for unseen entities by integrating latent type constraints and local subgraph reasoning through an LLM-guided framework. It introduces CATS, comprising two modules—Type-Aware Reasoning (TAR) and Subgraph Reasoning (SR)—and applies supervised fine-tuning with prompting to assess triple plausibility from both type-consistency and subgraph evidence perspectives. Through extensive experiments on WN18RR, FB15k-237, and NELL-995, CATS (full) achieves state-of-the-art performance in 16 of 18 settings, with an average MRR improvement of 7.2% and Hits@1 improvement of 10.1%, validating the value of combining latent type information with contextual reasoning. The results demonstrate the practical potential of LLM-guided KG reasoning in inductive, transductive, and few-shot scenarios, and highlight the importance of carefully designed prompts and fine-tuning over raw LLM capabilities alone.

Abstract

Inductive knowledge graph completion (KGC) aims to predict missing triples with unseen entities. Recent works focus on modeling reasoning paths between the head and tail entity as direct supporting evidence. However, these methods depend heavily on the existence and quality of reasoning paths, which limits their general applicability in different scenarios. In addition, we observe that latent type constraints and neighboring facts inherent in KGs are also vital in inferring missing triples. To effectively utilize all useful information in KGs, we introduce CATS, a novel context-aware inductive KGC solution. With sufficient guidance from proper prompts and supervised fine-tuning, CATS activates the strong semantic understanding and reasoning capabilities of large language models to assess the existence of query triples, which consist of two modules. First, the type-aware reasoning module evaluates whether the candidate entity matches the latent entity type as required by the query relation. Then, the subgraph reasoning module selects relevant reasoning paths and neighboring facts, and evaluates their correlation to the query triple. Experiment results on three widely used datasets demonstrate that CATS significantly outperforms state-of-the-art methods in 16 out of 18 transductive, inductive, and few-shot settings with an average absolute MRR improvement of 7.2%.

Context-aware Inductive Knowledge Graph Completion with Latent Type Constraints and Subgraph Reasoning

TL;DR

The paper tackles inductive knowledge graph completion for unseen entities by integrating latent type constraints and local subgraph reasoning through an LLM-guided framework. It introduces CATS, comprising two modules—Type-Aware Reasoning (TAR) and Subgraph Reasoning (SR)—and applies supervised fine-tuning with prompting to assess triple plausibility from both type-consistency and subgraph evidence perspectives. Through extensive experiments on WN18RR, FB15k-237, and NELL-995, CATS (full) achieves state-of-the-art performance in 16 of 18 settings, with an average MRR improvement of 7.2% and Hits@1 improvement of 10.1%, validating the value of combining latent type information with contextual reasoning. The results demonstrate the practical potential of LLM-guided KG reasoning in inductive, transductive, and few-shot scenarios, and highlight the importance of carefully designed prompts and fine-tuning over raw LLM capabilities alone.

Abstract

Inductive knowledge graph completion (KGC) aims to predict missing triples with unseen entities. Recent works focus on modeling reasoning paths between the head and tail entity as direct supporting evidence. However, these methods depend heavily on the existence and quality of reasoning paths, which limits their general applicability in different scenarios. In addition, we observe that latent type constraints and neighboring facts inherent in KGs are also vital in inferring missing triples. To effectively utilize all useful information in KGs, we introduce CATS, a novel context-aware inductive KGC solution. With sufficient guidance from proper prompts and supervised fine-tuning, CATS activates the strong semantic understanding and reasoning capabilities of large language models to assess the existence of query triples, which consist of two modules. First, the type-aware reasoning module evaluates whether the candidate entity matches the latent entity type as required by the query relation. Then, the subgraph reasoning module selects relevant reasoning paths and neighboring facts, and evaluates their correlation to the query triple. Experiment results on three widely used datasets demonstrate that CATS significantly outperforms state-of-the-art methods in 16 out of 18 transductive, inductive, and few-shot settings with an average absolute MRR improvement of 7.2%.

Paper Structure

This paper contains 29 sections, 7 equations, 3 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Embedding-based methods.
  4. Graph neural network (GNN)-based methods.
  5. Path-based methods.
  6. Text-enhanced methods.
  7. Preliminaries
  8. Problem specification.
  9. Definition 1 (Inductive KGC):
  10. Type properties of entities.
  11. Reasoning paths.
  12. Definition 2 (Reasoning path):
  13. Methodology
  14. Type-aware Reasoning (TAR)
  15. Subgraph Reasoning (SR)
  16. ...and 14 more sections

Figures (3)

  • Figure 1: A typical scenario of inductive knowledge graph completion. The model needs to infer the nationality of the unseen entity "António Guterres".
  • Figure 2: The statistics on the existence of reasoning paths for the query triples in the FB15k-237 (inductive) dataset.
  • Figure 3: The end-to-end pipeline of the proposed CATS framework.