Knowledge Graph Embedding Model Based on Spiking Neural-like Graph Attention Network for Relation Prediction

Area of Science:

• Artificial Intelligence
• Machine Learning
• Natural Language Processing

Background:

• Knowledge graphs (KGs) are crucial for NLP but suffer from incompleteness, limiting their utility.
• Predicting missing relations in KGs is a significant research challenge.

Purpose of the Study:

• To propose GEGS, a novel KG embedding framework for enhanced scalability and expressiveness in relation prediction.
• To address the limitations of incomplete knowledge graphs.

Main Methods:

• GEGS integrates GAT-SNP (Graph Attention Network with Spiking Neural P mechanisms) to capture complex relational structures.
• A BiLSTM-SNP component is incorporated to mitigate information loss in long-range and sequential path features.

Main Results:

• GEGS achieved superior performance in link prediction tasks on benchmark datasets (Kinship, FB15k-237, WN18RR).
• The model demonstrated state-of-the-art results across multiple evaluation metrics, including Hits@10 and MRR.

Conclusions:

• GEGS offers an effective solution for knowledge graph completion by improving relation prediction.
• The framework's enhanced scalability and expressiveness pave the way for large-scale knowledge base applications.