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

Abstract

The South China Sea (SCS) presents a significant research challenge due to severe data fragmentation and semantic heterogeneity across disparate historical, legal, and geopolitical sources. Conventional linear research approaches often fail to systematically integrate these disparate records. This study develops and demonstrates an automated LLM-driven framework for constructing the South China Sea Knowledge Graph (SCS-KG). The three-stage process (extraction, normalization, and verification) converts heterogeneous textual sources into a coherent, machine-readable structure that integrates the region’s historical, cultural, and geopolitical dimensions. After verification, the SCS-KG comprises 59,836 entities and 652,018 triples. The resulting knowledge graph supports three analytical applications: (1) evidence-based query answering that synthesizes facts across centuries (e.g., from ancient textual records to modern legal declarations); (2) discovery of implicit, multi-hop relationships linking state-level governance and micro-level social practices; and (3) rapid, entity-centric profiling of complex geopolitical instruments. This LLM-based approach provides a replicable model for structured knowledge representation and enables integrated, evidence-based analysis in South China Sea studies.