Abstract: Robotic process automation requires the use of user interaction logs to design process models.However,user interaction log often contains multiple tasks,and the execution of different tasks is likely to interfere with each other.To address the above problems,an approach to multi-task user interaction log segmentation based on graph embedding was proposed.In this approach,the action context information was first integrated into the direct following graph.Afterwards,the vector representation of the behavior was extracted by graph embedding algorithm.Finally,the user interaction log was segmented and clustered according to the precursor and successor correlation of the actions.In particular,to improve the accuracy of the action embedding,a weight enhancement method was proposed to optimize the sampling process in graph embedding algorithm.The experiments showed that the proposed approach could effectively solve the problem of user interaction log segmentation for multi-task sequential and interleaved execution.

Key words: robotic process automation, user interaction logs, log segmentation, graph embedding

摘要： 机器人流程自动化需要借助用户交互日志设计流程模型。然而,真实场景中的用户交互日志往往包含多种任务,同时不同的任务执行过程也很可能存在相互干扰。针对上述问题,本文提出一种基于图嵌入的多任务用户交互日志分段方法。该方法首先将行为上下文信息融入直接跟随图中,然后利用图嵌入算法提取行为的向量表征,最后根据行为前驱、后继相关性对用户交互日志进行切分和聚类,从而达到自动分段的目的。特别地,为了提高行为向量表征的准确性,提出权重增强的方法来优化图嵌入算法中的采样过程。实验结果表明,所提方法可以有效地解决多任务顺序和交错执行场景用户交互日志分段问题。

关键词: 机器人流程自动化, 用户交互日志, 日志分段, 图嵌入