An enhanced framework for knowledge graph embedding based on negative sample analogical reasoning.

Area of Science:

• Artificial Intelligence
• Data Science
• Machine Learning

Background:

• Knowledge graph embedding (KGE) models are effective for link prediction but struggle with complex relationships and missing data.
• Existing analogical reasoning methods for KGE rely on positive samples, which are often scarce in real-world incomplete knowledge graphs.
• This scarcity leads to suboptimal performance in analogical inference and link prediction.

Purpose of the Study:

• To propose a novel enhanced framework, Ne_AnKGE, for knowledge graph embedding.
• To address the limitations of positive sample dependency in analogical reasoning for KGE.
• To improve the representation of complex relationships and enhance link prediction accuracy.

Main Methods:

• Developed Ne_AnKGE, a framework utilizing negative sample analogical reasoning.
• Integrated enhanced TransE and RotatE models within the Ne_AnKGE framework to leverage multiple KGE approaches.
• Applied negative sample analogical reasoning to mitigate the scarcity of similar positive samples.

Main Results:

• Ne_AnKGE effectively mitigates the scarcity of positive samples in analogical reasoning.
• The framework enhances the link prediction capabilities of basic KGE models.
• Experimental results on FB15K-237 and WN18RR datasets demonstrate Ne_AnKGE's competitive performance.

Conclusions:

• Ne_AnKGE offers a robust solution for knowledge graph embedding, particularly in scenarios with data imbalance and incompleteness.
• Negative analogical reasoning is a viable strategy to improve KGE performance.
• The integration of multiple enhanced KGE models within Ne_AnKGE boosts its ability to handle complex relationships.