Integração de modelos de linguagem e RAG na criação de chatbots oftalmológicos

Abstract

Accessibility to ophthalmological services is an important factor in determining eye health, being influenced by the socioeconomic status of individuals. To facilitate access to information about eye health, recent works in the field focus on using established private language models or those with finetuning, both approaches involving additional costs, whether financial, data base needs, or complexity. This study proposes the development of a chatbot using open-source language models and retrieval augmented generation (RAG) techniques. Three techniques were evaluated naive RAG, HYDE and Rewrite-Retrieve-Read. To evaluate the retrieved context and the generated response, ChatGPT was used as a critic through the Ragas framework. The results indicate that it is possible to surpass the baseline performance of GPT-3.5 with the proposed techniques, reducing costs and attesting to the viability of similar projects.