Embedding-Based Entity Alignment of Cross-Lingual Temporal Knowledge Graphs.

Area of Science:

• Knowledge Graph Construction
• Artificial Intelligence
• Natural Language Processing

Background:

• Existing entity alignment methods primarily address static knowledge graphs.
• Temporal characteristics of entity relationships and attributes are often overlooked, leading to alignment inaccuracies.
• Cross-lingual entity alignment in temporal knowledge graphs remains an underexplored research area.

Purpose of the Study:

• To propose a novel entity alignment method for cross-lingual temporal knowledge graphs.
• To address the limitations of static and non-temporal approaches in entity alignment.
• To improve the accuracy and reliability of matching entities across different languages and time dimensions.

Main Methods:

• Developed CTEA, a joint embedding model combining entity, relation, and attribute embeddings.
• Utilized Graph Convolutional Networks (GCN) and TransE for embedding generation.
• Integrated distance and similarity calculations to enhance cross-lingual alignment reliability.

Main Results:

• The CTEA model demonstrated improved performance in entity alignment tasks.
• Experimental results showed an approximate 0.8–2.4 percentage point increase in Hits@m and MRR compared to state-of-the-art methods.
• The proposed method effectively handles temporal dynamics and cross-lingual complexities in knowledge graphs.

Conclusions:

• CTEA offers a robust solution for entity alignment in cross-lingual temporal knowledge graphs.
• The joint embedding approach and integrated similarity measures enhance alignment precision.
• This work contributes to advancing the field of knowledge graph completion and management.