Abstract: In business process modeling and analysis tasks,activities are usually represented as discrete symbols,which may results in the loss of activity semantics.To tackle this problem,a novel methodology for representation learning for activities named Activity to Vector(Activity 2Vec) was proposed to represent activities in event logs as real-valued vectors in a distributed semantic space,which could provide the support for modern deep learning techniques in business process modeling and analysis tasks.To address the limited-volume and sparsity issue of event logs,the raw log was enhanced through simulation of multiple process models.Furthermore,a vector representations learning algorithm was developed based on negative sampling,which was capable of learning both activity vectors and temporal interval vectors from the enhanced event log.Experiments on MIMIC III dataset demonstrated that the proposed method outperformed traditional Word2Vec in discovering high-quality activity vectors.

Key words: temporal activity, event log, representation learning, process tree, vector representations learning algorithm

摘要： 传统业务流程建模与分析任务中通常将活动表示为离散符号,损失了一定的语义信息。针对这一问题,提出了时序活动表示学习方法,使用多维实数向量对活动语义进行量化表示,为深度学习等现代人工智能技术在业务流程建模与分析中的应用提供基础支持。首先利用过程模型对事件日志的高层次抽象能力,通过过程模型挖掘及仿真对原始事件日志进行增强,扩大事件日志规模并强化活动关系统计信息;然后设计了融合活动关系和执行时间信息的向量表示学习算法,从增强后的事件日志中学习活动向量表示。通过在一个公开的真实医院诊疗日志语料上开展的实验研究验证了所提方法相比于传统的词向量学习方法具有明显优势。

关键词: 活动, 事件日志, 表示学习, 流程树, 向量表示学习算法