关键词: 事件流预测, 行为向量编码, 在线, 协同过滤, 业务流程监控

Abstract: Event stream prediction is the process of analyzing known process behavior to anticipate unknown states so that business activities are executed in the desired manner.The existing online event stream prediction methods mainly focus on the state of the event stream itself,and less considers the specific behavioral relationships between the event streams.To effectively analyze these behavioral relationships and incorporate them into the prediction process,an online event stream collaborative filtering recommendation algorithm based on event popularity as a vector was proposed to predict the next event.The structural similarity and behavioral similarity between event streams were first analyzed to determine the behavioral profile relationships of event streams.On this basis,the behavioral dependencies of the event streams were captured and the event streams were constructed into behavioral vectors.Then the traditional collaborative filtering recommendation algorithm was adapted to analyze the online event streams to achieve effective prediction of the next event stream.The relevant algorithms were implemented in the Pm4py framework and simulated in synthetic logs and real logs for prediction.The experimental results showed that the behavior vectors could reflect the behavioral relationships of the event streams and improve the effectiveness of the prediction.

Key words: event stream prediction, behavioral vector encoding, online, collaborative filtering, business process monitor