Convolutional Models with Multi-Feature Fusion for Effective Link Prediction in Knowledge Graph Embedding.

Area of Science:

• Artificial Intelligence
• Data Science
• Machine Learning

Background:

• Link prediction is crucial for knowledge graph embedding (KGE) but faces challenges with computational cost and complex relationships.
• Existing methods struggle to efficiently capture multifaceted relationships within knowledge graphs.

Purpose of the Study:

• To develop a more efficient and adept KGE link prediction methodology.
• To address the computational overhead and complexity constraints of current approaches.

Main Methods:

• Amalgamation of convolutional operators with graph structural information.
• Integration of entity and relational neighbor information for enhanced convolutional models.
• Inclusion of edge-specific data into the convolutional model's input for customizable architecture and parameters.

Main Results:

• The proposed method significantly enhances performance over existing convolution-based link prediction benchmarks.
• Superior results were observed on the FB15k, WN18, and YAGO3-10 datasets.
• The approach demonstrates improved efficiency and adeptness in KGE link prediction.

Conclusions:

• The novel approach effectively addresses limitations in current KGE link prediction techniques.
• The methodology offers a flexible and powerful tool for predicting relationships in real-world knowledge graphs.
• This research paves the way for more efficient and capable knowledge graph embedding.