Complex Embedding with Type Constraints for Link Prediction.

Area of Science:

• Artificial Intelligence
• Data Science
• Knowledge Representation

Background:

• Knowledge graphs store entities and relations, offering ontology-based information.
• Type constraints within knowledge graphs are crucial for accurate link prediction.
• Existing methods often struggle to fully leverage type constraints for enhanced prediction.

Purpose of the Study:

• To propose a novel complex embedding method, CHolE, for improved link prediction in large-scale knowledge graphs.
• To integrate type constraints into complex representational embeddings.
• To enhance the accuracy of predicting relationships between entities.

Main Methods:

• Introduced complex circular correlation to extend real-valued compositional representations to complex domains.
• Developed CHolE with two components: a type constraint model and a relation learning model.
• Utilized modulus constraints and captured interactions in modulus and phase angles of complex embeddings.

Main Results:

• CHolE demonstrated superior performance compared to previous state-of-the-art methods on benchmark datasets.
• The integration of type constraints significantly improved the model's link prediction performance.
• The model accurately acquired entity relatedness by leveraging complex embedding properties.

Conclusions:

• CHolE offers an effective approach for link prediction in knowledge graphs by incorporating type constraints.
• Complex embeddings provide a richer representation for capturing entity relationships and constraints.
• The proposed method advances the field of knowledge graph representation learning.