During the execution of business processes, data loss in event logs due to information system failures or manual recording errors and so on, resulting in the generation of the missing event logs, the use of such missing logs can seriously affect the quality of business process analysis results. Aiming at the problem of repairing missing logs, most of the existing research focus on data or behavioral perspective, and there is little work on repairing events log from the perspective of integrating data and behavior. Therefore, this paper proposed a multi-perspective event log repair method based on the BERT model. This method trained the model using a two-layer BERT model from the perspective of data and behavior fusion, and captured the bidirectional semantic information and long-term dependencies of input attributes by the BERT model's pre-training tasks(masked attribute model(MAM) and masked event model(MEM) ) and the attention mechanism of the Transformer encoding block. Then it used fine-tuning strategies for training the model to repair missing values in the event log in a predictive manner. Finally, it carried out evaluation and analysis on publicly available datasets, and the results showed that the proposed method performs well in repairing event logs.

[2023-08-11] Accepted Paper /doc/2023.06.0253/v1

[2024-02-05] Printed Article /doc/2023.06.0253/v2