Розроблення підходу ДО створення спеціалізованого словника для тренування чат-ботів з генеративним штучним інтелектом

This paper investigates the process that builds a subject-specific training dictionary for training a chatbot involving generative artificial intelligence. The task addressed is to reproduce the structured vocabulary characteristic of the relevant subject area from subject-specific knowledge when interacting with a chatbot. The result of this study is the construction of a model for the process that sequentially manages independent user requests. The model made it possible to estimate the mathematical expectation of the stage number at which the processing of the request by the chatbot is completed. Based on the constructed mathematical model, linear and logical-probabilistic models for building a specialized dictionary have been proposed. The linear model searches for a combination of words by sequentially searching for terms. The result of this approach is the comparison of a keyword from the query with the corresponding term or word form from the dictionary. The logical-probabilistic model is based on the target cell – a probable word from the user's query. This is explained by the possibility of defining a word that agrees with the term of the XML dictionary and has maximum relevance to the user query. A methodology and algorithm for building a specialized dictionary have been suggested. The tests made it possible to obtain average signature values for the response at an error of 0.004%, as well as ensure the stability of the results. In practice, this could be used under the conditions of forming a probability distribution of possible word combinations for generating a response. The proposed approach could be used in practical tasks of chatbots' domain adaptation, in particular at project support portals and in scientific libraries, as well as for improving intelligent dialog systems focused on the formation of refined user queries