Structured knowledge representation of the South China Sea: An LLM-based knowledge graph approach

Area of Science:

• Geospatial Intelligence
• Computational Social Science
• Information Science

Background:

• The South China Sea (SCS) research is hindered by fragmented and heterogeneous data from historical, legal, and geopolitical sources.
• Traditional linear research methods struggle to integrate these disparate records effectively.

Purpose of the Study:

• To develop and demonstrate an automated Large Language Model (LLM)-driven framework for constructing the South China Sea Knowledge Graph (SCS-KG).
• To convert heterogeneous textual sources into a coherent, machine-readable structure integrating the region's historical, cultural, and geopolitical dimensions.

Main Methods:

• A three-stage automated process: extraction, normalization, and verification of textual data.
• Utilizing LLMs to process and structure disparate information into a knowledge graph format.
• Verification stage to ensure data integrity and coherence within the SCS-KG.

Main Results:

• The constructed SCS-KG contains 59,836 entities and 652,018 triples.
• The knowledge graph enables evidence-based query answering across historical periods and modern declarations.
• Facilitates discovery of implicit, multi-hop relationships between governance and social practices.

Conclusions:

• The LLM-based SCS-KG framework offers a replicable model for structured knowledge representation.
• Enables integrated, evidence-based analysis for complex South China Sea studies.
• Supports advanced applications including query answering, relationship discovery, and entity profiling.