[2K4-IS-1a] Knowledge engineering

[2K4-IS-1a-01] Synthetic Data Generation Using GANs and LLM with Knowledge Graph: Application in Real-World Multi-Domain Datasets

〇Ke YU¹, Shigeru Ishikura², Yukari Usukura², Yuki Shigoku², Teruaki Hayashi¹ (1. the university of Tokyo, 2. Infomart Corporation)

Keywords:generative AI, synthetic data, knowledge graph

This study propose the application of generative AI to create high-quality synthetic datasets that resemble four distinct real-world datasets, each representing a different domain. The proposed approach leverages the strengths of generative models, such as LLMs and GANs, combined with domain-specific knowledge encoded in knowledge graphs. The study aims to maintain the logical consistency, statistical distribution, and structural characteristics of the original datasets while ensuring data quality and usability. The results demonstrate the potential of generative AI to provide scalable, privacy-preserving synthetic datasets for research and practical applications across various fields, including geography, transactions, customer demographics, and contractual data.