GraphGen: Enhancing Supervised Fine-Tuning for LLMs with Knowledge-Driven Synthetic Data Generation

2025-05-26

AI Summary

To address the scarcity, high cost, and poor long-tail coverage of high-quality annotated data in supervised fine-tuning, this paper proposes a knowledge graph (KG)-driven synthetic data generation framework tailored for atomic, aggregative, and multi-hop question answering. Methodologically, it introduces fine-grained KG modeling and an Expected Calibration Error (ECE)-based knowledge gap identification mechanism, integrated with multi-hop neighborhood sampling and style-controllable large language model (LLM) generation. This design significantly improves factual accuracy and long-tail knowledge coverage. Empirically, the approach outperforms existing synthetic data baselines across closed-book, knowledge-intensive tasks, enhancing both model generalization and factual consistency. The framework’s code and datasets are publicly released.

Technology Category

• Natural Language Processing: Code Generation / Program Synthesis from Natural Language
• Data Mining & Knowledge Management: Linked Open Data, Knowledge Graphs & KB Completion
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Agentic search

Abstract

Fine-tuning for large language models (LLMs) typically requires substantial amounts of high-quality supervised data, which is both costly and labor-intensive to acquire. While synthetic data generation has emerged as a promising solution, existing approaches frequently suffer from factual inaccuracies, insufficient long-tail coverage, simplistic knowledge structures, and homogenized outputs. To address these challenges, we introduce GraphGen, a knowledge graph-guided framework designed for three key question-answering (QA) scenarios: atomic QA, aggregated QA, and multi-hop QA. It begins by constructing a fine-grained knowledge graph from the source text. It then identifies knowledge gaps in LLMs using the expected calibration error metric, prioritizing the generation of QA pairs that target high-value, long-tail knowledge. Furthermore, GraphGen incorporates multi-hop neighborhood sampling to capture complex relational information and employs style-controlled generation to diversify the resulting QA data. Experimental results on knowledge-intensive tasks under closed-book settings demonstrate that GraphGen outperforms conventional synthetic data methods, offering a more reliable and comprehensive solution to the data scarcity challenge in supervised fine-tuning. The code and data are publicly available at https://github.com/open-sciencelab/GraphGen.

Problem

• Addressing costly high-quality data needs for LLM fine-tuning
• Overcoming synthetic data inaccuracies and knowledge gaps
• Enhancing QA data diversity and relational complexity

Innovation

• Knowledge graph-guided synthetic data generation
• Multi-hop sampling for complex relations
• Style-controlled generation for diverse QA data