DESIGN OF GENERATION SYSTEM BASED ON ARTIFICIAL INTELLIGENCE ALGORITHM. PART 2

Artificial intelligence is often understood as a specific algorithm or an ensemble of machine learning models that enable solving applied problems of various kinds, most often characterized by inherent nonlinearity, while considering the specifics of the research domain and the possibility of applying non-trivial approaches. One such task is the generation of new objects or combinations. This study examines the full development cycle of a generation system, from data collection to the final model incorporating artificial intelligence elements. Typically, the models and algorithms used require an initial dataset that facilitates training and pattern discovery and can later be applied to other research areas. The first part of the article is dedicated to solving the problem of dataset formation. Methods for protecting online sources were explored, a parser for automatic data collection from electronic sources and resources was developed, preprocessing of the obtained information was performed, and a dataset of word combinations was compiled. This is the most labor-intensive and crucial stage in the development of the generation system, as all subsequent results—their quality and reliability—depend entirely on the completeness, cleanliness, and correctness of the structured data. The second part focuses on the next stages of system development, specifically data visualization, tuning of mathematical model parameters, and validation of the generated results. During dataset formation, the information is represented in a tabular format, making visual analysis an essential step. In this study, embedding vectors obtained through a simple Word2Vec neural network are used for visualization, as the data is presented in textual form. These vectors are transformed into a two-dimensional space using the t-SNE method. At the final stage of system development, a classical machine learning algorithm the Markov Chain is employed as the generation model. This approach allows for the creation of a relatively simple yet effective system that fully aligns with the modeling process for discovering new patterns and combinations.