Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Efficient Relational Context Perception for Knowledge Graph Completion

Wenkai Tu, Guojia Wan, Zhengchun Shang, Bo Du

TL;DR

Knowledge Graph Completion seeks to infer missing facts in incomplete graphs. The paper introduces Triple Receptance Perception (TRP), a sequential encoder that yields dynamic, context-sensitive embeddings, paired with a Tucker decomposition decoder for robust scoring. By employing reciprocal learning and 1-N scoring, the method achieves state-of-the-art or competitive results in link prediction and triple classification across benchmarks, while maintaining parameter efficiency and interpretable embedding structures. This approach advances practical KG reasoning by capturing fine-grained context and interactions with reduced computational cost.

Abstract

Knowledge Graphs (KGs) provide a structured representation of knowledge but often suffer from challenges of incompleteness. To address this, link prediction or knowledge graph completion (KGC) aims to infer missing new facts based on existing facts in KGs. Previous knowledge graph embedding models are limited in their ability to capture expressive features, especially when compared to deeper, multi-layer models. These approaches also assign a single static embedding to each entity and relation, disregarding the fact that entities and relations can exhibit different behaviors in varying graph contexts. Due to complex context over a fact triple of a KG, existing methods have to leverage complex non-linear context encoder, like transformer, to project entity and relation into low dimensional representations, resulting in high computation cost. To overcome these limitations, we propose Triple Receptance Perception (TRP) architecture to model sequential information, enabling the learning of dynamic context of entities and relations. Then we use tensor decomposition to calculate triple scores, providing robust relational decoding capabilities. This integration allows for more expressive representations. Experiments on benchmark datasets such as YAGO3-10, UMLS, FB15k, and FB13 in link prediction and triple classification tasks demonstrate that our method performs better than several state-of-the-art models, proving the effectiveness of the integration.

Efficient Relational Context Perception for Knowledge Graph Completion

TL;DR

Knowledge Graph Completion seeks to infer missing facts in incomplete graphs. The paper introduces Triple Receptance Perception (TRP), a sequential encoder that yields dynamic, context-sensitive embeddings, paired with a Tucker decomposition decoder for robust scoring. By employing reciprocal learning and 1-N scoring, the method achieves state-of-the-art or competitive results in link prediction and triple classification across benchmarks, while maintaining parameter efficiency and interpretable embedding structures. This approach advances practical KG reasoning by capturing fine-grained context and interactions with reduced computational cost.

Abstract

Knowledge Graphs (KGs) provide a structured representation of knowledge but often suffer from challenges of incompleteness. To address this, link prediction or knowledge graph completion (KGC) aims to infer missing new facts based on existing facts in KGs. Previous knowledge graph embedding models are limited in their ability to capture expressive features, especially when compared to deeper, multi-layer models. These approaches also assign a single static embedding to each entity and relation, disregarding the fact that entities and relations can exhibit different behaviors in varying graph contexts. Due to complex context over a fact triple of a KG, existing methods have to leverage complex non-linear context encoder, like transformer, to project entity and relation into low dimensional representations, resulting in high computation cost. To overcome these limitations, we propose Triple Receptance Perception (TRP) architecture to model sequential information, enabling the learning of dynamic context of entities and relations. Then we use tensor decomposition to calculate triple scores, providing robust relational decoding capabilities. This integration allows for more expressive representations. Experiments on benchmark datasets such as YAGO3-10, UMLS, FB15k, and FB13 in link prediction and triple classification tasks demonstrate that our method performs better than several state-of-the-art models, proving the effectiveness of the integration.
Paper Structure (24 sections, 14 equations, 6 figures, 5 tables)

This paper contains 24 sections, 14 equations, 6 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methodology
  4. Encoder
  5. Triple Receptance Perception Architecture
  6. Time Mixing Module
  7. Channel Mixing Module
  8. Integration of Modules
  9. Decoder
  10. Training
  11. Experiments
  12. Experiment setup
  13. Datasets
  14. Baseline
  15. Evaluation Metrics
  16. ...and 9 more sections

Figures (6)

  • Figure 1: Illustration of a Knowledge Graph with a potential link prediction task: inferring the missing triple (Cristiano Ronaldo, CountryLiveIn, Saudi Arabia) based on the existing facts
  • Figure 2: Architecture of our method.
  • Figure 3: Effect of the number of layers (Left) and embedding dimension (Right) on MRR for FB15k and YAGO3-10
  • Figure 4: Training time per epoch across dataset proportion on YAGO3-10 (Left) and parameter efficiency comparison between TRP and Transformer across different embedding dimensions. (Right)
  • Figure 5: t-SNE visualizations of entity embeddings from YAGO3-10 and FB15k. Each point represents an entity embedding, with different colors indicating various categories
  • ...and 1 more figures