Parsing Large XES Files for Discovering Process Models: A Big Data Problem

Abstract

Process mining is a group of techniques for retrieving de-facto models using system traces. Discovering algorithms can obtain mathematical models exploiting the information contained into list of events of activities. Completeness of the traces is relevant for the accuracy of the final results. Noiseless traces appear as an ideal scenario. The performance of the algorithms is significant reduce if the log files are not processed efficiently. XES is a logical model for process logs stored in data centric xml files. In real processes the sizes of the logs increase exponentially. Parsing XES files is presented as a big data problem in real scenarios with dense traces. Lazy parsers and DOM models are not enough appropriate in scenarios with large volumes of data. We discuss this problematic and how to use indexing techniques for retrieving useful information for process mining. An XES compression schema is also discussed for reducing the index construction time.