TECNOLOGÍAS DEL LENGUAJE PARA TUTORIZACIÓN DOCENTE CON AGENTES CONVERSACIONALES
Fecha
2024-07-11
Autores
Título de la revista
ISSN de la revista
Título del volumen
Editor
Jaén: Universidad de Jaén
Resumen
Este trabajo tiene como objetivo potenciar y modernizar la educación a través de las tecnologías del
lenguaje, específicamente utilizando modelos del lenguaje respaldados por sistemas RAG. Para ello, se
exploran diversas técnicas para mejorar estos modelos y se elige emplear un sistema RAG junto con
LLaMa 2, un modelo del lenguaje seleccionado por varias razones que benefician significativamente
este proyecto.
Durante el desarrollo del trabajo, se elabora una base de conocimiento para una asignatura piloto, en
este caso, ‘Fundamentos de Bases de Datos’. Para ello, se realiza web scraping y se utilizan librerías
como LangChain, ChromaDB, Transformers y Torch. Además, se implementa un modelo de creación de
embeddings y se aplican técnicas de ingeniería de prompts.
Finalmente, se lleva a cabo un estudio de los resultados obtenidos, se presentan conclusiones y se
proponen posibles mejoras para el futuro.
This work aims to enhance and modernise education through language technologies, specifically using language models supported by RAG systems. To this end, various techniques are explored to enhance these models and a RAG system is chosen to be used in conjunction with LLaMa 2, a language model selected for a number of reasons that significantly benefit this project. During the development of the work, a knowledge base is elaborated for a pilot subject, in this case, 'Fundamentals of Databases'. For this purpose, web scraping is performed and libraries such as LangChain, ChromaDB, Transformers and Torch are used. In addition, an embedding creation model is implemented and prompts engineering techniques are applied. Finally, a study of the results obtained is carried out, conclusions are presented and possible improvements for the future are proposed.
This work aims to enhance and modernise education through language technologies, specifically using language models supported by RAG systems. To this end, various techniques are explored to enhance these models and a RAG system is chosen to be used in conjunction with LLaMa 2, a language model selected for a number of reasons that significantly benefit this project. During the development of the work, a knowledge base is elaborated for a pilot subject, in this case, 'Fundamentals of Databases'. For this purpose, web scraping is performed and libraries such as LangChain, ChromaDB, Transformers and Torch are used. In addition, an embedding creation model is implemented and prompts engineering techniques are applied. Finally, a study of the results obtained is carried out, conclusions are presented and possible improvements for the future are proposed.