Knowledge Graph Alignment Under Scarce Supervision: A General Framework With Active Cross-View Contrastive Learning.

Area of Science:

• Artificial Intelligence
• Data Science
• Knowledge Representation

Background:

• Knowledge graphs (KGs) are essential for data integration but rarely complete.
• Knowledge Graph Alignment (KGA) enhances KG coverage but existing methods often neglect relations and require extensive labeled data.

Purpose of the Study:

• To propose a general framework for simultaneous entity and relation alignment in KGs.
• To address the limitations of scarce supervision and the neglect of relations in traditional KGA.

Main Methods:

• Developed a framework with two core components: relation-enhanced active instance selection (RAS) and cross-view contrastive learning (CCL).
• RAS guides the selection of valuable instances for labeling using relational information.
• CCL leverages contrastive learning across different views to amplify limited supervision signals.

Main Results:

• The proposed framework consistently improves KG alignment performance across various KG pairs under scarce supervision.
• Both RAS and CCL components contribute to the enhanced alignment accuracy.
• The framework is model-agnostic, adaptable to existing entity and relation alignment techniques.

Conclusions:

• The novel framework effectively enhances KG coverage by simultaneously aligning entities and relations.
• The approach significantly boosts alignment performance even with limited labeled data.
• This work offers a viable solution for practical KG alignment challenges.