Abstract: Business process modeling recommendation is one of the main ways to assist process modeling. Process similarity is the core technology of this method, which directly affects the accuracy of process recommendation. Process similarity includes two aspects：Graph structure similarity and behavior similarity. In order to solve the problem of difficulty in quantifying process similarity and low accuracy, a single-layer neural network-based process similarity quantification method is proposed. Firstly, this paper builds a behavior-based process system, establishes business processes, independent paths, and function expression relationships between process nodes. Secondly, it uses one-hot coding to represent the process nodes with initialization vectors, and establishes a single-layer neural network model to record the process trajectory of log data as the training set, the process nodes are mapped to a low-dimensional continuous vector space, the vector of each process node is obtained., and the vector form of the business process is obtained. Finally the similarity between the processes is calculated by the cosine similarity. The accuracy of the proposed method is verified through experiments, and the application of the method in the field of process recommendation is introduced.

Key words: business process, process behavior, process similarity, process recommendation, artificial intelligence, neural network

摘要： 业务流程建模推荐是辅助流程建模的主要方式之一，流程相似性是该方法的核心技术，直接影响流程推荐的准确率。流程相似性包括流程图结构相似性与行为相似性两方面，为了解决流程相似性量化困难，准确率低的问题，提出基于单层神经网络的流程相似度的量化方法。对构建基于行为的流程体系，建立业务流程、独立路径、流程节点之间的函数表达关系；使用独热编码对流程节点进行初始化向量表示，建立单层神经网络模型，以记录流程运行轨迹的日志数据为训练集，将流程节点映射到低维连续的向量空间，获得每个流程节点的向量，进而得到业务流程的向量形式；通过余弦相似度计算流程间的相似性。通过实验验证所提方法的准确率，并介绍了该方法在流程推荐领域的应用方式。

关键词: 业务流程, 流程行为, 流程相似性, 流程推荐, 人工智能, 神经网络