Development of an approach to chat-bot personalization with generative artificial intelligence when realize an online assistant

The object of research is the interaction in the “human – machine” system during the user's interaction with generative artificial intelligence. The relevance of the research topic is due to the need to provide assistance to users in a narrow professional topic. To implement the goal set in the work, a model of operator decomposition was developed using the “Goals, Objects, Methods, and Selection rules” GOMS technology, taking into account the multi-level cognitive functions of a person. For this purpose, microoperators were used, which are responsible for combining various actions to find an answer to a question. A model with the decomposition of the operator μ was developed, which is responsible for cognitive functions when creating a request during human interaction with a chatbot based on artificial intelligence. The work used interaction with the ChatGPT chatbot. The proposed decomposition algorithm was used as the basis for the online assistant plugin. The implementation is made in JavaScript, which allows it to be used on any sites and portals. The main components of the plugin are the interface for entering a query, a multi-level search mechanism on the site and in connected specialized libraries. The API integration of the plugin with ChatGPT was implemented. As a result of the work, a study was conducted to experimentally determine the values of action and movement operators that are related to human mental activity and algorithmized in the online assistant. According to the results of the experiment, it was taken into account that for a chatbot, queries using foreign language signs and symbols and queries in the user's usual natural language are equivalent. To communicate with ChatGPT using the plugin, it is necessary to adhere to uniqueness and clarity when forming narrowly professional queries. The result was obtained that when querying in natural language on a topic familiar to the user, the online assistant adapts to the requirements more slowly. But at the same time, the speed of finding an answer and its formulation is accelerated. The problem of personalizing the online assistant was solved. This became possible thanks to the analysis of user behavior through the detailing of the query by micro-operators in the GOMS model. This allows to personalize the online assistant without user registration, only based on its behavior when forming a request. The proposed approach can be used to create online assistants for the implementation of highly specialized complex projects on web platforms.