A lightweight CNN-based knowledge graph embedding model with channel attention for link prediction.

Area of Science:

• Artificial Intelligence
• Data Science
• Machine Learning

Background:

• Knowledge graph embedding (KGE) represents entities and relations in a continuous vector space.
• Link prediction (LP) is a key KGE application for inferring missing facts.
• Enhancing feature interactions in KGE improves semantic representation for LP.

Purpose of the Study:

• To propose IntSE, a novel lightweight CNN-based KGE model.
• To improve KGE performance for link prediction through enhanced feature interactions.
• To leverage channel attention for adaptive feature recalibration in KGE.

Main Methods:

• Developed IntSE, a CNN-based KGE model incorporating efficient CNN components.
• Integrated a channel attention mechanism to adaptively recalibrate feature responses.
• Evaluated IntSE on public datasets for link prediction tasks.

Main Results:

• IntSE demonstrated superior performance compared to existing CNN-based KGE models.
• The model effectively increased feature interactions between entity and relation embeddings.
• Channel attention mechanism successfully enhanced relevant features for improved LP accuracy.

Conclusions:

• IntSE offers an effective and lightweight approach for knowledge graph embedding.
• The proposed model significantly advances the state-of-the-art in link prediction.
• IntSE's architecture provides a robust framework for capturing complex semantic relationships in KGs.