Parallel Mining Small Patterns from Business Process Traces

Abstract

Hadoop MapReduce has arrived to solve the problem of treatment of big data, also the parallel treatment, with this framework the authors analyze, process a large size of data. It based for distributing the work in two big steps, the map and the reduce steps in a cluster or big set of machines. They apply the MapReduce framework to solve some problems in the domain of process mining how provides a bridge between data mining and business process analysis, this technique consists to mine lot of information from the process traces; In process mining, there are two steps, correlation definition and the process inference. The work consists in first time of mining patterns whom are the work flow of the process from execution traces, those patterns present the work or the history of each party of the process, the authors' small patterns are represented in this work by finite state automaton or their regular expression, the authors have only two patterns to facilitate the process, the general presentation of the process is the combination of the small mining patterns. The patterns are represented by the regular expressions (ab)* and (ab*c)*. Secondly, they compute the patterns, and combine them using the Hadoop MapReduce framework, in this work they have two general steps, first the Map step, they mine small patterns or small models from business process, and the second is the combination of models as reduce step. The authors use the business process of two web applications, the SKYPE, and VIBER applications. The general result shown that the parallel distributed process by using the Hadoop MapReduce framework is scalable, and minimizes the execution time.