Дослідження з інтеграції моделей штучного інтелекту в системи підтримки клієнтів

Abstract

The subject of this work is dedicated to studying fine-tuning techniques of large language models within the context of custom domain. The goal of the work is to analyze methods of fine-tuning large language models, compare their productivity and quality of results with limited amount of training data, and explore their utilization in customer support systems. This article describes preparing a dataset, various methods of fine-tuning and evaluates them using a custom data set. Domain specific help articles and the client queries in a "question-answer" format were used as data.