Automating Implementation of Business Logic of Multi Subject-Domain IS on the Base of Machine Learning, Data Programming and Ontology-Based Generation of Labeling Functions

Abstract

The article analyzes the possibility of using machine learning (ML) models built with the use of a data programming approach for the implementation of the business logic of multi subject-domain information systems (MSIS). MSIS characterized by the heterogeneity and variability of users, data and functions. ML considered as a promising approach to the implementation of business logic in conditions of variability and heterogeneity of MSIS. To effectively use ML models within MSIS the model-centric architecture on the base of data programming approach is proposed. To quickly obtain high-quality sets of labelled training data, it is proposed to use the data programming approach which based on simple code snippets named labelling functions. To ensure the reuse of knowledge and accelerate the implementation of business logic based on ML models, a technology for automated generation of labelling functions based on domain ontologies has been proposed. The performed experiments confirmed the efficiency and potential effectiveness of the proposed technology.