SCMEA: A stacked co-enhanced model for entity alignment based on multi-aspect information fusion and bidirectional...

Area of Science:

• Artificial Intelligence
• Knowledge Representation and Reasoning

Background:

• Entity alignment is crucial for knowledge graph completion and fusion.
• Existing methods using knowledge representation learning have limitations in mining hidden information.

Purpose of the Study:

• To propose SCMEA, a novel cross-lingual entity alignment framework.
• To enhance entity alignment by effectively mining multi-aspect information and improving entity representations.

Main Methods:

• SCMEA employs diverse representation learning models for multi-aspect information embedding.
• An adaptive weighted mechanism unifies embeddings, overcoming missing and non-uniform information.
• A stacked relation-entity co-enhanced model with Global Entity Attention refines entity representations.
• Improved bidirectional contrastive learning optimizes parameters and mitigates the hubness problem.

Main Results:

• SCMEA demonstrates significant performance improvements in cross-lingual entity alignment.
• Ablation studies and experiments on five datasets validate the model's effectiveness and robustness.
• The framework successfully addresses challenges like missing information and the hubness problem.

Conclusions:

• SCMEA offers a robust and effective solution for cross-lingual entity alignment.
• The proposed multi-aspect information fusion and bidirectional contrastive learning significantly advance the field.
• The framework shows strong potential for knowledge graph completion and fusion applications.