Decomposed and parallel process discovery: A framework and application

The rapid growth of event data motivates the developing of decomposed and parallel process discovery, which solves process discovery from a large event log by decomposing the log into multiple small event logs, discovering process models from these small logs in parallel, and merging discovered process models. As such, process discovery from a large event log can be solved in less time. Currently, the passage and maximal decomposition based techniques are successful in the activity partition by decomposing causal graph structure derived from a large event log. In this paper, we propose a five-step framework, based on which we can build various decomposed and parallel process discovery techniques by simply combining and adapting existing techniques. Then, we propose a technique, RPSTHD, based on the framework using the refined process structure tree (RPST), heuristic miner, process mining using integer linear programming (ILP), etc. An experimental evaluation shows that our technique significantly outperforms the state-of-the-art decomposed discovery techniques in both efficiency and effectiveness.