On the Software Implementation of Cognitive Interoperable Agent-Based Systems

Abstract

Purpose of research. The purpose of the work is to develop recommendations for the software implementation of cognitive agent-based systems that ensure interoperability in the interaction of software cognitive agents with different properties. A software implementation that determines semantic proximity based on machine learning can automatically and quickly highlight important key concepts and find associations, simplifying and speeding up the process of working with text data during a dialogue between agents, one of which is a human. The proposed approach is based on the assumption that computer systems can perform some “anthropomorphic” functions, similar to human ability to think.Methods. Domain knowledge is determined by training an artificial neural network. To indicate the semantics of remarks and other information, it is proposed to use tagging and determining the semantic proximity of key phrases from speeches presented in written form.Results. The system was implemented in the Python programming language. The Word2Vec model with Skip-gram architecture was used as a neural network model for text vectorization. For training, two text sets with information about computer science and zoology were used. Based on the results of comparing texts on two topics, one can judge the performance of the system to determine the semantic proximity of textual information.Conclusion. The subsystem for determining the semantic proximity of text information based on machine learning technologies, which forms the basis for the software implementation of cognitive interoperable agent-based systems, will improve the efficiency of existing or developed applications that involve a large amount of text information.