DynaSynth: a large language model fine-tuning data dynamic synthesis agent tool for unstructured cybersecurity corpus

Guan Yongjian; Piao Chengkai; Wang Buhong; Zhao Bofu; Li Siqi; Zhao Zhengyang

doi:10.11959/j.issn.2096-109x.25249

您当前的位置：

首页 >

文章列表页 >

DynaSynth: a large language model fine-tuning data dynamic synthesis agent tool for unstructured cybersecurity corpus

Papers | 更新时间：2026-05-11

- DynaSynth: a large language model fine-tuning data dynamic synthesis agent tool for unstructured cybersecurity corpus
- Chinese Journal of Network and Information Security Vol. 12, Issue 2, Pages: 156-168(2026)
- 作者机构：
  
  空军工程大学信息与导航学院，陕西西安 710077
- 作者简介：
- 基金信息：
  
  The National Natural Science Foundation of China(62472437)
- DOI：10.11959/j.issn.2096-109x.25249
  CLC： TP393
- Received：27 December 2025，
  
  Revised：2026-03-29，
  
  Accepted：30 March 2026，
  
  Published：25 April 2026
- 稿件说明：
移动端阅览
关永健, 朴乘锴, 王布宏, 等. DynaSynth：面向非结构化网络安全语料的大语言模型微调数据动态生成代理工具[J]. 网络与信息安全学报, 2026, 12(2): 156-168.

Guan Yongjian, Piao Chengkai, Wang Buhong, et al. DynaSynth: a large language model fine-tuning data dynamic synthesis agent tool for unstructured cybersecurity corpus[J]. Chinese Journal of Network and Information Security, 2026, 12(2): 156-168.
关永健, 朴乘锴, 王布宏, 等. DynaSynth：面向非结构化网络安全语料的大语言模型微调数据动态生成代理工具[J]. 网络与信息安全学报, 2026, 12(2): 156-168. DOI： 10.11959/j.issn.2096-109x.25249.

Guan Yongjian, Piao Chengkai, Wang Buhong, et al. DynaSynth: a large language model fine-tuning data dynamic synthesis agent tool for unstructured cybersecurity corpus[J]. Chinese Journal of Network and Information Security, 2026, 12(2): 156-168. DOI： 10.11959/j.issn.2096-109x.25249.

摘要

随着模型训练技术和数据规模的迅速发展，大语言模型已经能够胜任各种通用任务，但在网络安全垂直领域，大语言模型的发展仍受数据瓶颈的制约。为解决这一难题，提出了面向非结构化语料的大语言模型微调数据动态生成代理工具DynaSynth。该工具首先利用向量嵌入辅助模型将原始文本切割为文本块，并转化为可检索增强的向量数据；然后通过对文档进行受众－体裁分析，指导生成具有多样化问答对风格且兼具通用性的数据。在此过程中，用户可以通过可视化界面实时查看生成的数据，并借助提示工程对生成策略进行优化调整，确保生成内容精准契合实际需求。实验结果表明，使用DynaSynth生成的微调数据能够显著提高大语言模型在垂直领域任务中的性能。该工具不仅具备高效的数据生成能力，还能够有效探索数据的多样化生成路径。

Abstract

With the rapid development of model training technologies and data scale

large language models have been able to handle various general tasks. However

in the vertical field of cybersecurity

the development of large language models is still constrained by data bottlenecks. To solve this problem

a data dynamic generation agent tool named DynaSynth for large language models based on unstructured text corpora was proposed. This tool first uses vector embeddings to assist the model in splitting the original text into blocks and converting them into retrievable enhanced vector data; then

through audience and genre analysis of the documents

it guides the generation of data with diverse question-answer pair styles and universality. During this process

users can view the generated data in real time through a visual interface and optimize the generation strategy by leveraging prompt engineering to ensure that the generated content precisely meets the actual needs. Experimental results show that the fine-tuning data generated by DynaSynth can significantly improve the performance of large language models in vertical domain tasks. This tool not only has an efficient data generation capability

but also effectively explores the diverse generation paths of data.

关键词

Keywords

references

Chang Y P , Wang X , Wang J D , et al . A survey on evaluation of large language models [J ] . ACM Transactions on Intelligent Systems and Technology , 2024 , 15 ( 3 ): 1 - 45 .

Wei J , Tay Y , Bommasani R , et al . Emergent abilities of large language models [PP ] . arXiv ( 2022-10-26 )[ 2025-12-27 ] . arXiv:arXiv. 2206 . 07682 .

Wang L , Ma C , Feng X Y , et al . A survey on large language model based autonomous agents [J ] . Frontiers of Computer Science , 2024 , 18 ( 6 ): 186345 .

Guo T C , Chen X Y , Wang Y Q , et al . Large language model based multi-agents: a survey of progress and challenges [PP ] . arXiv ( 2024-04-19 )[ 2025-12-27 ] . arXiv:arXiv. 2402.01680.

Chen W Z , Koenig S , Dilkina B . RePrompt: planning by automatic prompt engineering for large language models agents [PP ] . arXiv ( 2025-02-13 )[ 2025-12-27 ] . arXiv:arXiv. 2406 . 11132 .

Khatami S , Frantz C . Prompt engineering guidance for conceptual agent-based model extraction using large language models [PP ] . arXiv ( 2024-12-05 )[ 2025-12-27 ] . arXiv:arXiv. 2412 . 04056 .

Sun M J , Han R J , Jiang B Y , et al . LAMBDA: a large model based data agent [J ] . Journal of the American Statistical Association , 2025 : 1 - 13 .

Qiao B , Li L Q , Zhang X , et al . TaskWeaver: a code-first agent framework [PP ] . arXiv ( 2024-06-20 )[ 2025-12-27 ] . arXiv:arXiv. 2311.17541.

Wu Q , Bansal G , Zhang J , et al . AutoGen: enabling next-Gen LLM applications via multi-agent conversations [PP ] . arXiv ( 2023-08-16 )[ 2025-12-27 ] . arXiv:arXiv. 2308.08155.

Qian C , Liu W , Liu H Z , et al . ChatDev: communicative agents for software development [PP ] . arXiv ( 2024-06-05 )[ 2025-12-27 ] . arXiv:arXiv. 2307.07924.

Canto Á B D , Blázquez G M , Lajarín A P , et al . Distilabel: an AI feedback (AIF) framework for building datasets with and for LLMs [EB ] . 2026 .

Chen Z H , Jiang W L , Li J Z , et al . GraphGen: enhancing supervised fine-tuning for LLMs with knowledge-driven synthetic data generation [PP ] . arXiv ( 2025-05-26 )[ 2025-12-27 ] . arXiv:arXiv. 2505.20416.

Chen D Y , Huang Y L , Ma Z J , et al . Data-juicer: a one-stop data processing system for large language models [C ] // Proceedings of the Companion of the 2024 International Conference on Management of Data . New York : ACM , 2024 : 120 - 134 .

Lewis P , Perez E , Piktus A , et al . Retrieval-augmented generation for knowledge-intensive NLP tasks [C ] // Proceedings of the Advances in Neural Information Processing Systems 33: Annual Conference on Neural Information Processing Systems 2020 . San Diego : NeurIPS , 2020 : 9459 - 9474 .

Gao Y F , Xiong Y , Gao X Y , et al . Retrieval-augmented generation for large language models: a survey [PP ] . arXiv ( 2024-03-27 )[ 2025-12-27 ] . arXiv:arXiv. 2312.10997.

刘文彦 , 霍树民 , 仝青 , 等 . 网络安全评估与分析模型研究 [J ] . 网络与信息安全学报 , 2018 , 4 ( 4 ): 1 - 11 .

Liu W Y , Huo S M , Tong Q , et al . Research on models of network security evaluation and analysis [J ] . Chinese Journal of Network and Information Security , 2018 , 4 ( 4 ): 1 - 11 .

Lála J , O’Donoghue O , Shtedritski A , et al . PaperQA: retrieval-augmented generative agent for scientific research [PP ] . arXiv [ 2025-12-27 ] . arXiv:arXiv. 2312.07559.

Hao X T , Zhu R J , Zhang G , et al . Reformulation for pretraining data augmentation [PP ] . arXiv ( 2023-12-14 )[ 2025-12-27 ] . arXiv:arXiv. 2502 . 04235 .

Long L , Wang R , Xiao R X , et al . On LLMs-driven synthetic data generation, curation, and evaluation: a survey [PP ] . arXiv ( 2025-05-19 )[ 2025-12-27 ] . arXiv:arXiv. 2406.15126.

张繁 , 谢凡 , 江颉 . 网络威胁安全数据可视化综述 [J ] . 网络与信息安全学报 , 2018 , 4 ( 2 ): 34 - 39 .

Zhang F , Xie F , Jiang J . Survey on the visualization technologies of threatening security data in cyber space [J ] . Chinese Journal of Network and Information Security , 2018 , 4 ( 2 ): 34 - 39 .

Yu Y C , Chiang T H , Tsai C W , et al . Primus: a pioneering collection of open-source datasets for cybersecurity LLM training [PP ] . arXiv ( 2025-10-03 )[ 2025-12-27 ] . arXiv:arXiv. 2502.11191.

Du Y H , Oraby S , Perera V , et al . Schema-guided natural language generation [PP ] . arXiv ( 2020-11-04 ) [ 2025-12-27 ] .arXiv:arXiv. 2005 . 05480 .

Stureborg R , Alikaniotis D , Suhara Y . Characterizing the confidence of large language model-based automatic evaluation metrics [C ] // Proceedings of the 18th Conference of the European Chapter of the Association for Computational Linguistics . Stroudsburg : ACL , 2024 : 76 - 89 .

Luo Y F , Tang Y M , Shen C F , et al . Prompt engineering through the lens of optimal control [PP ] . arXiv ( 2023-11-03 )[ 2025-12-27 ] . arXiv:arXiv. 2310.14201.

Marcondes F S , Gala A , Magalhães R , et al . Using ollama [M ] // Natural Language Analytics with Generative Large-Language Models: A Practical Approach with Ollama and Open-Source LLMs . Cham : Springer Nature Switzerland , 2025 : 23 - 35 .

Abid A , Abdalla A , Abid A , et al . Gradio: hassle-free sharing and testing of ML models in the wild [PP ] . arXiv ( 2019-06-06 )[ 2025-12-27 ] . arXiv:arXiv. 1906 . 02569 .

Yang A , Li A F , Yang B S , et al . Qwen3 technical report [PP ] . arXiv ( 2025-05-14 )[ 2025-12-27 ] . arXiv:arXiv. 2505.09388.

Krishna V B . AttackQA: development and adoption of a dataset for assisting cybersecurity operations using fine-tuned and open-source LLMs [PP ] . arXiv ( 2024-11-01 )[ 2025-12-27 ] . arXiv:arXiv. 2411 . 01073 .

Zheng L , Chiang W L , Sheng Y , et al . Judging LLM-as-a-judge with MT-Bench and Chatbot arena [J ] . Proceedings of the 37th International Conference on Neural Information Processing Systems. NewYork: NeurIPS , 2023 : 46595 - 46623 .

Papineni K , Roukos S , Ward T , et al . BLEU: a method for automatic evaluation of machine translation [C ] // Proceedings of the 40th Annual Meeting on Association for Computational Linguistics . Morristown : ACL , 2002 : 311 - 318 .

Views

下载量

CSCD

Alert me when the article has been cited

提交

Tools

Publicity Resources

Data augmentation based multi-view contrastive learning graph anomaly detection

Pseudo-siamese network image tampering localization model based on reinforced samples

Insider threat detection based on operational attention and data augmentation

Research on synthesis data generation method for logo recognition

Related Author

XU Li

DAI Yuanfei

LIN Xingpeng

LI Jiayin

LI Yifan

Jinwei WANG

Bin MA

Xiangyang LUO

Related Institution

College of Computer and Cyber Security, Fujian Normal University

Fujian Provincial Key Laboratory of Network Security and Cryptology

College of Computer and Information Engineering, Nanjing University of Technology

Engineering Research Center of Digital Forensics, Ministry of Education, Nanjing University of Information Science and Technology

Department of Computer Science, Nanjing University of Information Science and Technology

⁰