A Framework for Dynamic Knowledge Graph Ontology Incremental Alignment based on Large Language Model Agents
Year: 2026
Type: preprint
Abstract: Dynamic knowledge graphs used in risk management and compliance monitoring evolve continuously, yet most ontology alignment methods still operate on static snapshots, which creates a persistent trade-off between semantic precision and update latency. To address this limitation, we propose DOKA-Agent, a large language model agent framework for dynamic ontology incremental alignment. The framework follows a perceive-plan-act workflow: it captures structural and content deltas from a graph, enriches the semantic signature of affected entities using contextual information, retrieves candidates through a hybrid dense-sparse strategy, and then produces relation-aware alignment decisions with confidence estimation and constraint checking. Experiments on DBP15K dynamic partitions show that DOKA-Agent improves Hit@1, Hit@10, and F1 in both static and dynamic settings while substantially reducing alignment time. These results indicate that agentic orchestration provides a practical way to maintain semantic consistency in continuously evolving knowledge graphs.
Source: Research Square
Authors: Hao Qin, Fusheng Wei, Bin Liang, Wobin Lin
Institution: Guangdong Power Grid Company (China)
Language: English
Topic: Semantic Web and Ontologies
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7167225579
