Creating a Language Module for Predicting a Patient's Preliminary Diagnosis for Interaction through a Dialog Agent

Abstract

In recent years, conversational agents based on artificial intelligence are considered as a promising method of primary interaction with a patient when he visits a clinic. Indeed, this helps to relieve the registry, as well as optimize the flow of patients within the medical institution. It is worth noting that the customer experience that individuals have gained through the active use of the online environment to solve everyday tasks motivates them to implement dialog agents in such a way that the interaction is quick and convenient. It is due to this that motivation is created for the further use of this communication channel. However, chatbots in Russian currently do a poor job of interpreting text data received from the user. Thus, it is a difficult task to isolate symptoms from a user-entered string, and in connection with this there is a further problem of text classification.. The paper is devoted to the description of the process of designing and developing a language model for predicting a patient's preliminary diagnosis when interacting using a dialog agent. A priori knowledge about diseases and their symptoms, semi-structured data from disease general discussion forums, and a generated set of texts using ChatGPT were selected as data for model training and algorithm validation. The article describes the general idea of the created library, reveals the topic of disease classification, and analyzes the quality metrics of the developed models.