Uma abordagem para planejamento instrucional apoiada em processos de decisão Markovianos parcialmente observáveis

Abstract

One of the major difficulties encountered in a tutoring session is the choice of appropriate pedagogical actions that take into account the student’s previous individual knowledge and aims to lead the student to achieve a proficiency in the assimilation of a set of concepts. These actions should consider the student’s profile and adapt the teaching according to his or her learning style and preferences in order to keep the student engaged. Considering that personalized education has a positive influence on the assimilation of the disciplinary content, the tendency is that the new virtual learning environments are to be implemented in a way that adapts to the student’s profile. The elaboration of an adaptive system should consider the inclusion of a module responsible for generating an instructional action plan, capable of modifying its own initial planning to diagnose failures in the comprehension of essential fundamentals of content as they are perceived throughout the tutoring session. Therefore, it is the responsibility of the adaptive system to conduct a teaching plan considering all the difficulties that are observed for each student in particular. To achieve this goal, this system should have the capacity to infer the student’s knowledge, propose actions and observe the evolution of learning, fully supervising the learning process. Given the difficulties presented, this paper presents the proposal and the implementation of an instructional module for a support system for education based on Partially Observable Markov Decision Processes. The choice of the model is justified by its capacity to make decisions in the face of the uncertainties that characterize the environment: uncertainty regarding the knowledge the learner presents about the concepts to be assimilated and the non-deterministic effects of the proposed actions. The problem of the sequencing of actions produced by this module can be represented as a problem of probabilistic planning and are solved by planning algorithms. There are several probabilistic planning algorithms that could be used in the development of an instructional module for an adaptive teaching system, and the results obtained by each of them may differ. A planner or planning system is software that implements a planning algorithm. In this work, a methodology based on machine learning is applied to select a probabilistic planner that provides the choice of the best instructional actions. The choice of the planner is accomplished by considering the characteristics of a specification domain that teaches concepts and evaluates them according to the student’s proficiency. At the end of the selection process, the result was validated by executing the planner indicated by the methodology to the domain that describes the instructional module. The performance obtained by its execution was compared with that of other candidate planners, thus proving the efficiency in the application of the proposed method.