Abstract: To improve the automation of business process model construction in projects or enterprises and effectively utilize unstructured data within processes,a method for automated business process modeling based on Large Language Models (LLMs) was proposed.In this modeling framework,a Large Language Model was employed to construct an input collection module that accepted inputs in natural language text and other unstructured formats.A simple markup language was used to intermediate represent the business processes.Sequential transformation rules were constructed to fine-tune the Large Language Model's learning on the usage of the markup language,facilitating the conversion from natural language inputs to intermediate representation.Additionally,to correct grammatical errors in results and enhance modeling outcomes,a multi-agent correction module was introduced to automatically inspect and rectify results.Finally,modeling experiments were conducted on business processes with different characteristics,validating the automation of the proposed modeling method and its effectiveness in extracting implicit loop structures.

Key words: business process modeling, automated modeling, large language models, multi-agent, loop structure recognition

摘要： 为了提高项目或企业在业务流程模型构建过程的自动化程度并有效利用流程的非结构数据,提出一种基于大语言模型(LLMs)的业务流程自动建模方法。在该建模方法框架中,采用大语言模型构建输入采集模块,接受自然语言文本等非结构化形式输入;以一种简单标记语言对业务流程进行中间表示,通过构造序列转换规则数据集对大语言模型进行微调,使其学习如何使用该标记语言,完成自然语言输入到中间表示的转换过程。为了纠正结果的语法错误并改善建模结果,引入多智能体纠错模块对结果进行自动检查与修正。最后,针对具有不同特点的业务流程进行建模实验,验证了所提建模方法的自动性以及在提取隐性循环结构方面的有效性。

关键词: 业务流程建模, 自动化建模, 大语言模型, 多智能体, 循环结构识别