Abstract: A reasonable encoding technique can greatly improve the accuracy of remaining time prediction.Therefore,five types of event encoding techniques were designed.All the events contained in the business process were extracted,and the event encoding techniques was used to encode the events.According to the sequential characteristics of business processes,different types of remaining time prediction models were constructed,and the event encoding vector was used as the input of the prediction model,so as to evaluate the impact of event encoding methods on the remaining time prediction.Experiments on eight public event logs demonstrated that the GloVe encoding technique achieved the best accuracy compared to other techniques in terms of process remaining time prediction.The obtained experiment results provided valuable insights and guidance to researchers and practitioners to choose the most appropriate event encoding technique to achieve the best remaining time prediction accuracy.

Key words: business process, remaining time prediction, deep learning, event encoding techniques

摘要： 合理的事件编码方式有助于提升业务流程剩余时间预测效果,为此,有针对性地设计出5种事件编码方式。首先,抽取业务流程包含的全部事件,并利用事件编码方式对获取的事件进行编码。其次,根据业务流程序列性的特点,构建不同类型的剩余时间预测模型,同时将事件编码向量作为预测模型的输入,从而评估事件编码方式对业务流程剩余时间预测的影响。在8个公开事件日志数据集上进行实验,结果表明GloVe事件编码方式在提高业务流程剩余时间预测效果上是最有效的。该实验结果可帮助研究者和从业者选择最合适的事件编码方式以实现最佳剩余时间预测效果。

关键词: 业务流程, 剩余时间预测, 深度学习, 事件编码方式