ERPA i.e. project 2020-1.1.2-PIACI-KFI-2020-00165 is a project carried out at the Institutes of Informatics and Mathematics of the Faculty of Mechanical Engineering and Informatics of the University of Miskolc.
One of the main means of increasing efficiency in administrative and IT processes is the implementation of process automation as widely as possible. The purpose of robotic process automation (RPA, Robotic Process Automation) is to automate business processes at a high level by creating and running software robots (bots). Such robots perform high-volume, repetitive tasks with greater accuracy and efficiency than humans. RPA tools are designed to automate routine tasks that contain structured data and have deterministic results, and the steps to be performed are governed by rules. In addition, an important characteristic of RPA tools is their ability to perform user interactions together with the associated data management operations. Robotic Process Mining (RPM) aims to explore automatable processes with machine learning tools. The development of process mining methods is still a largely unexplored problem area. One of the most promising approaches is event discovery based on activity logs.
The research team working on the project had to cover several main tasks. On the one hand, you had to examine the extent to which each event log model format is suitable for highlighting the required information. The second subproblem was the development of the flexible log preprocessing framework, which brings log files of different formats into a single object structure. The third step was the development of the classical, automata-based method of exploring event graph schemes. In the process, new functions were added to the basic procedure, significantly increasing its efficiency. Since an additional goal of the research was the analysis of the neural network-based schema discovery, here we first performed the evaluation of the neural networks predicting the classical sequence of events. Based on the results, the MLP and LSTM networks were selected for further investigations. In a further phase of the research, a new neural network architecture and processing model was developed, which is already suitable for the detection of XOR and AND scheme branches. In addition, in the framework of the project, in order to prepare the research for the next period, we also performed analyzes regarding the applicability of the GAN network type and the NLP-based topic identification of documents containing Hungarian text. The sample systems developed in the framework of the project were developed in Python.
The project also provided an opportunity to build an effectively collaborative team, which brings together representatives of several different fields of expertise. Members of the group:
The cooperation of the representatives of the different fields of expertise revealed many interesting approaches.