Modification and completion of geological structure knowledge graph based on pattern matching.

Area of Science:

• Computer Science
• Artificial Intelligence
• Geology

Background:

• Knowledge graphs are crucial for intelligent systems but lack methods for mining domain-specific patterns.
• Existing research primarily focuses on information retrieval, neglecting the physical meaning of subgraph structures.

Purpose of the Study:

• To propose a novel method and framework for knowledge graph pattern mining using graph attention networks (GAT).
• To address the challenge of identifying specific subgraph structures with physical significance in domain-specific knowledge graphs.

Main Methods:

• Transformed domain-specific patterns into subgraph structures incorporating topological information and entity attributes.
• Framed pattern mining as an approximate subgraph matching problem, treating patterns as query graphs and knowledge graphs as data graphs.
• Utilized an improved relational graph attention network with an adaptive edge deletion mechanism for subgraph matching.

Main Results:

• Successfully implemented end-to-end training for the proposed method.
• Achieved approximate subgraph matching on an existing dataset.
• Demonstrated the method's effectiveness in mining key patterns within complex geological structure knowledge graphs.

Conclusions:

• The developed GAT-based framework effectively mines meaningful patterns in complex knowledge graphs.
• The approach facilitates the discovery of structurally and attribute-wise significant subgraphs.
• This work advances knowledge graph pattern mining, particularly for specialized applications like geological structure analysis.