Diseño de un chatbot utilizando técnicas de procesamiento de lenguaje natural para soporte de servicios de información de una IES

Abstract

In the current IT career landscape, the integration of innovative technologies has become a pillar to improve the quality of services offered to students and the general public, this is how the need for the development of a chatbot for IT in a Facebook social profile was born as a prototype to provide assistance with academic information of the career without time restriction. In terms of development technologies, there are several types of chatbots, from those based on rules to those driven by AI and deep learning. However, the predominant trend is the use of technologies such as ChatGPT, which are based on generative language models and NLP to provide natural and contextual responses, making them very popular and novel. The proposal focused on the development of the chatbot prototype driven by the long language model GPT-3.5 Turbo, the purpose was to optimize the way users interact with the chatbot and thus obtain the effective response by streamlining interactions and providing accurate responses and uninterrupted availability. The implementation of natural language processing techniques facilitated the contextualization of questions in the chatbot, being crucial for academic information management support. To strengthen this approach, conducting systematic research proved to be an essential step as it contributed to the understanding of similar cases and guided the optimal selection of techniques and tools in the development of the chatbot. In the prototype, technologies such as GPT-3.5, Pinecone and Langchain were employed, creating a specialized system that delivers accurate responses. Facebook integration was crucial for thorough testing and error management, improving the algorithm's efficiency. After evaluation, the chatbot achieved a solid 94.5% accuracy, supporting its effectiveness on the platform