Semantic information-based attention mapping network for few-shot knowledge graph completion.

Area of Science:

• Artificial Intelligence
• Data Science
• Natural Language Processing

Background:

• Few-shot Knowledge Graph Completion (FKGC) infers new facts using limited data.
• Current FKGC methods underutilize textual semantic information, relying heavily on structural data.
• This limits the accuracy and applicability of existing FKGC models.

Purpose of the Study:

• To propose an innovative model, the Semantic Information-based Attention Mapping Network (SI-AMN), for FKGC.
• To effectively integrate both structural and textual semantic information for enhanced completion accuracy.
• To address the limitations of existing FKGC methods in leveraging textual data.

Main Methods:

• Developed a Semantic Information-based Attention Mapping Network (SI-AMN).
• Incorporated a dual-information fusion mechanism combining structural and textual data.
• Utilized a semantic encoder for textual feature extraction and an attention mapping network for semantic interaction learning.

Main Results:

• SI-AMN achieved a 40% improvement in prediction accuracy on benchmark datasets.
• Demonstrated superior performance compared to state-of-the-art FKGC methods.
• Ablation studies confirmed the effectiveness of individual model components.

Conclusions:

• The proposed SI-AMN model significantly enhances knowledge graph completion accuracy.
• Highlights the critical importance of integrating semantic information into graph completion tasks.
• Opens new avenues for future research in leveraging rich textual data for knowledge graph completion.