Approximate Query on Temporal Knowledge Graphs via Two-Level Embeddings

Area of Science:

• Computer Science
• Artificial Intelligence
• Data Science

Background:

• Approximate querying is crucial for knowledge graphs (KGs) in real-world applications.
• Existing methods primarily address static KGs, neglecting their dynamic nature.
• Temporal knowledge graphs (TKGs) incorporate time-evolving information, posing unique challenges.

Purpose of the Study:

• To develop an effective method for approximate querying in temporal knowledge graphs (TKGs).
• To address the limitations of existing methods in handling dynamic and time-aware data.
• To improve the accuracy and relevance of query results from evolving KGs.

Main Methods:

• Proposed a Two-Level Approximate Query (TLAQ) method for TKGs.
• Enhanced graph convolutional network (GCN) eigenmatrix for improved vertex and graph embeddings.
• Introduced relational reliability and attributive confidence at the vertex level.
• Unified timestamp encoding at the graph level to strengthen the embedding model.

Main Results:

• The TLAQ method demonstrated effectiveness in handling approximate queries on TKGs.
• The proposed approach showed improved performance compared to existing methods in experimental evaluations.
• The two-level embedding strategy successfully captured temporal dynamics and relationships.

Conclusions:

• The TLAQ method offers a robust solution for approximate querying in dynamic TKGs.
• Enhancing graph embeddings with temporal information is key to improving query accuracy.
• This work contributes to more efficient and effective information retrieval from evolving knowledge bases.