Tuck-KGC: based on tensor decomposition for diabetes knowledge graph completion model integrating Chinese and Western...

Area of Science:

• Medical Informatics
• Artificial Intelligence
• Biomedical Data Science

Background:

• Medical knowledge graphs are crucial for advanced medical applications like personalized diagnostics and therapies.
• Incompleteness in medical knowledge graphs, due to missing entities or relationships, hinders their effectiveness.
• Existing tensor decomposition methods for knowledge graph completion often neglect vital entity type information, leading to inaccurate predictions.

Purpose of the Study:

• To address the incompleteness and inaccuracy issues in medical knowledge graphs.
• To propose a novel method, Tucker Decomposition Knowledge Graph Completion (Tuck-KGC), that integrates entity type information.
• To enhance the accuracy of link prediction in medical knowledge graphs.

Main Methods:

• Developed the Tucker Decomposition Knowledge Graph Completion (Tuck-KGC) model.
• Incorporated medical entity types into the tensor decomposition framework by mapping types to vectors.
• Integrated these type vectors into the knowledge graph representation learning process.
• Created the Dia dataset, a comprehensive medical knowledge graph for precision analysis.

Main Results:

• The Tuck-KGC model demonstrated improved link prediction accuracy.
• Experimental results showed an approximate 8% increase in link prediction accuracy compared to baseline methods.
• The incorporation of entity type information significantly enhanced the model's ability to predict accurate relationships.

Conclusions:

• The Tuck-KGC method effectively leverages entity type information to improve medical knowledge graph completion.
• Integrating entity types enhances the accuracy and reliability of predicted relationships, crucial for intelligent medical services.
• The Dia dataset provides a valuable resource for evaluating and advancing medical knowledge graph research.