Knowledge graph embedding with shared latent semantic units.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Data Science

Background:

• Knowledge Graph Embedding (KGE) models project entities and relations into a low-dimensional space.
• Existing KGE models often struggle with data sparsity, where most entities and relations appear infrequently.
• This sparsity limits the effectiveness of current KGE approaches.

Purpose of the Study:

• To propose a general technique for knowledge transfer in KGE to overcome data sparsity.
• To enhance the representational power of KGE models by enabling information sharing.
• To improve the performance of KGE models on knowledge graphs with sparse data.

Main Methods:

• Introduced Latent Semantic Units (LSUs) as sub-components of entity and relation embeddings.
• Developed a method for semantically similar entities/relations to share LSUs.
• Enabled knowledge transfer through shared LSUs among related entities and relations.

Main Results:

• The proposed technique successfully enhances existing KGE models.
• Demonstrated improved representations of knowledge graphs through knowledge transfer.
• Showcased the effectiveness of LSUs in mitigating data sparsity issues in KGE.

Conclusions:

• Latent Semantic Units provide a general and effective method for knowledge transfer in KGE.
• The LSU approach significantly improves KGE model performance, especially in sparse data scenarios.
• This technique offers a promising direction for building more robust and accurate KGE models.