Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Multi-level Shared Knowledge Guided Learning for Knowledge Graph Completion

Yongxue Shan, Jie Zhou, Jie Peng, Xin Zhou, Jiaqian Yin, Xiaodong Wang

TL;DR

The paper addresses knowledge graph completion by exploiting two levels of shared knowledge: dataset-level shared features revealed through text-based expansion of triples, and task-level knowledge sharing via a dynamic, balanced multi-task learning framework. It introduces SKG-KGC, a bi-encoder-based architecture that separates encoding of known elements from missing ones, augmented with dataset expansion and a dynamically weighted loss over head/entity prediction and relation prediction subtasks. Empirical results on WN18RR, FB15k-237, and Wikidata5M show competitive or state-of-the-art performance, with clear ablations confirming the contributions of dataset expansion, bi-encoder structure, and balanced multi-task learning; case studies illustrate improved handling of lexically similar candidates and inductive reasoning. Overall, SKG-KGC demonstrates that explicit, multi-level shared knowledge can significantly enhance KGC, offering a scalable path to integrating textual and structural information in knowledge graphs.

Abstract

In the task of Knowledge Graph Completion (KGC), the existing datasets and their inherent subtasks carry a wealth of shared knowledge that can be utilized to enhance the representation of knowledge triplets and overall performance. However, no current studies specifically address the shared knowledge within KGC. To bridge this gap, we introduce a multi-level Shared Knowledge Guided learning method (SKG) that operates at both the dataset and task levels. On the dataset level, SKG-KGC broadens the original dataset by identifying shared features within entity sets via text summarization. On the task level, for the three typical KGC subtasks - head entity prediction, relation prediction, and tail entity prediction - we present an innovative multi-task learning architecture with dynamically adjusted loss weights. This approach allows the model to focus on more challenging and underperforming tasks, effectively mitigating the imbalance of knowledge sharing among subtasks. Experimental results demonstrate that SKG-KGC outperforms existing text-based methods significantly on three well-known datasets, with the most notable improvement on WN18RR.

Multi-level Shared Knowledge Guided Learning for Knowledge Graph Completion

TL;DR

The paper addresses knowledge graph completion by exploiting two levels of shared knowledge: dataset-level shared features revealed through text-based expansion of triples, and task-level knowledge sharing via a dynamic, balanced multi-task learning framework. It introduces SKG-KGC, a bi-encoder-based architecture that separates encoding of known elements from missing ones, augmented with dataset expansion and a dynamically weighted loss over head/entity prediction and relation prediction subtasks. Empirical results on WN18RR, FB15k-237, and Wikidata5M show competitive or state-of-the-art performance, with clear ablations confirming the contributions of dataset expansion, bi-encoder structure, and balanced multi-task learning; case studies illustrate improved handling of lexically similar candidates and inductive reasoning. Overall, SKG-KGC demonstrates that explicit, multi-level shared knowledge can significantly enhance KGC, offering a scalable path to integrating textual and structural information in knowledge graphs.

Abstract

In the task of Knowledge Graph Completion (KGC), the existing datasets and their inherent subtasks carry a wealth of shared knowledge that can be utilized to enhance the representation of knowledge triplets and overall performance. However, no current studies specifically address the shared knowledge within KGC. To bridge this gap, we introduce a multi-level Shared Knowledge Guided learning method (SKG) that operates at both the dataset and task levels. On the dataset level, SKG-KGC broadens the original dataset by identifying shared features within entity sets via text summarization. On the task level, for the three typical KGC subtasks - head entity prediction, relation prediction, and tail entity prediction - we present an innovative multi-task learning architecture with dynamically adjusted loss weights. This approach allows the model to focus on more challenging and underperforming tasks, effectively mitigating the imbalance of knowledge sharing among subtasks. Experimental results demonstrate that SKG-KGC outperforms existing text-based methods significantly on three well-known datasets, with the most notable improvement on WN18RR.
Paper Structure (19 sections, 9 equations, 3 figures, 11 tables)

This paper contains 19 sections, 9 equations, 3 figures, 11 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Method
  4. Model Structure
  5. Dataset Expansion
  6. Bi-Encoder Architecture
  7. Balanced Multi-Task Learning
  8. Training
  9. Inference
  10. Experiments
  11. Experimental Setup
  12. Main Results
  13. Ablation Studies
  14. Further Exploration of Dataset-level Knowledge
  15. Further Exploration of Balanced Multi-Task Learning
  16. ...and 4 more sections

Figures (3)

  • Figure 1: (a) The greater the proportion of triples sharing the same $(r,t)$ or $(h,r)$, the more accessible knowledge we acquire. (b) The average number of connected entities in many-to-one and one-to-many relations indicates the imbalanced distribution between head entities and tail entities, respectively.
  • Figure 2: An overview of the SKG-KGC model.
  • Figure 3: The impact of the number of sentences on performance metrics.