COMPLEX EMBEDDINGS OF BUSINESS PROCESSES IN THE CLASSIFICATION PROBLEM

Abstract

Background. Optimization of business processes is an important problem in the management of organizational systems. The article is dedicated to the problem of creating business processes’ embeddings that allow to efficiently analyze the logs of information system on the progress of organizational tasks by machine learning methods. The purpose of the work is to study the analysis and optimization of business processes, based on a vector representation of the main parameters of organizational processes. The application of machine learning methods and process mining will efficiently solve the problems of classifying business processes considering their content, complexity and labor intensity. Materials and methods. Business process analysis methods based on information systems logs (process mining), classical machine learning methods and neural network technologies, organizational systems management methods are used. Results. In the course of the work, an overview of existing methods for creating document embeddings from the natural language processing and graphs analysis was carried out, their applicability to the creation of embedding vectors of business processes was evaluated. A number of experiments have been conducted to compare the effectiveness of using Bag-of-words, Tf-idf, Trace2vec, Graph2vec methods in the vectorization problem. Conclusions. The results of the experiments showed the effectiveness of the trace2vec method for short and medium-length processes and the graph2vec method for longlength processes.

Keywords

• vector representation of the process
• vectorization methods
• analysis of business process
• machine learning
• process mining
• graph analysis
• neural network