关键词: 剩余时间预测, 业务流程, 双向循环神经网络, 注意力机制

Abstract: Most of existing deep learning approaches to business process remaining time prediction are used the traditional Long-Short Term Memory (LSTM) neural networks.Due to the limited capacity of modeling temporal sequential data,there is still much room for improvement in predicting accuracy.For the deficiency,the business process remaining time prediction method based on leverage bidirectional recurrent neural networks was proposed.An attention mechanism to automatically learn the weights of each event was also introduced in a business process instances.To tackle the variety among business process instances of different lengths,a transfer learning mechanism to learn prediction models for ongoing process instances was designed with a particular length.Extensive experiments showed that the proposed approach outperformed existing process-model-based approaches and LSTM-based approaches.

Key words: remaining time prediction, business process, bidirectional recurrent neural networks, attention mechanism