Adaptación dinámica al usuario en un sistema de enseñanza mediante aprendizaje por refuerzo

Abstract

Uno de los problemas más importantes de los sistemas de educación a distancia es personalizar la enseñanza a cada estudiante, adaptando su política pedagógica dependiendo de las necesidades de aprendizaje que tengan los estudiantes. Los Sistemas de Educación Adaptativos e Inteligentes en Web (del inglés Web-based Adaptive and Intelligent Educational Systems) (SEAIs en Web) son sistemas de educación basados en Internet donde se aplican técnicas de inteligencia artificial con el objetivo de adaptar el contenido del sistema a los estudiantes. Definir políticas pedagógicas efectivas en estos sistemas es uno de los principales problemas de los SEAIs en Web, decidiendo qué, cómo y cuándo mostrar el contenido del curso a los estudiantes. En el trabajo realizado en la presente tesis doctoral se propone definir el problema de soporte adaptativo a la navegación a través del contenido del sistema y de presentación de dicho contenido como un problema de Aprendizaje por Refuerzo. Al aplicar el modelo de aprendizaje por refuerzo en el módulo pedagógico del sistema, éste será capaz de aprender automáticamente la mejor política pedagógica para cada estudiante individualmente, basados únicamente en la experiencia adquirida con otros estudiantes de características de aprendizajes similares, como hace el tutor en las aulas.

___________________________________________________

Last years, distance educational systems have been improved opening new perspectives on different ways to teach. The use of Internet as a tool for the educational systems helps the students and avoid the physical barriers in the access to the classrooms and the incompatibilities due to the different students system platforms. One of the most important issues in distance educational systems is to personalize the teaching to each student, adapting its pedagogical policy according to the pedagogical student needs. The application of artificial intelligence techniques to educational systems allows the systems to adapt in an intelligent way to the students, representing implicitly or explicitly the pedagogical policies, abilities and expert knowledge. The Web-based Adaptive and Intelligent Educational Systems (Web-based AIES) use artificial intelligence techniques in order to adapt the content to the students according to their pedagogical needs. One of the most important issues in these systems is to define effective pedagogical strategies for tutoring students according to their needs. The pedagogical strategies define what, how and when to show the system content to the students. In this PhD. Thesis we propose to use a pedagogical knowledge representation based on the Reinforcement Learning (RL) model. Using this model, the system is able to automatically provide adaptive navigation support and presentation support to the students (choosing the best presentation format for the content). The system learns which is the best pedagogical way to teach each student individually based only on acquired experience with other students with similar learning characteristics, like a human tutor does. In this dissertation we show how the definition of the pedagogical policies in the AIESs can be considered as a Reinforcement Learning problem from a theoretical point of view. Next, we show how the system is able to learn to teach from a theoretical point of view, using simulated students. Finally, we show how the system is able to teach in a practical point of view, interacting with real students of our University through Internet.