Active learning simulators (ALSs) allow students to practice and carry out experiments in a safe environment – anytime, anywhere. Well-designed simulations may enhance learning, and provide the bridge from concept to practical understanding. Nevertheless, learning with ALS depends largely on the student’s ability to explore and interpret the performed experiments. By adding an Intelligent Tutoring System (ITS), it is possible to provide individualized personal guidance to students. The challenges are how an ITS properly assesses the cognitive state of the student based on the results of experiments and the student’s interaction, and how it provides adaptive feedback to the student. In this chapter we describe how an ITS based on Dynamic Decision Networks (DDNs) is applied in an undergraduate Physics scenario where the aim is to adapt the learning experience to suit the learners’ needs. We propose employing Probabilistic Relational Models (PRMs) to facilitate the construction of the model. These are frameworks that enable the definition of Probabilistic Graphical and Entity Relationship Models, starting from a domain, and in this case, environments of ALSs. With this representation, the tutor can be easily adapted to different experiments, domains, and student levels, thereby minimizing the development effort for building and integrating Intelligent Tutoring Systems (ITS) for ALSs. A discussion of the methodology is addressed, and preliminary results are presented.
TopIntroduction
Most Active Learning Simulators (ALSs) allow students (or users in general) to explore the effect that different variables have on the behavior of the system under a given scenario. These types of systems have also been called Open Learning Environments (OLEs). The main benefits of OLE systems are that they provide the student with the opportunity to learn through free exploration. The student has the freedom to modify different parameters and to observe their effects (Bunt, 2003). Nevertheless, students often repeat a given experiment several times without a clear understanding of the phenomenon behind the working scenario. For example, while interacting with OLEs some hyperactive students could try several options in the system, but the students’ learning cannot be ensured. Another student could explore diverse variables and observe their behavior during simulation. However the student may not be able to connect the theory with the phenomenon that he is observing (McHaney, 2002). Another student could make a selection of parameters that may be enable him to create a link between the theoretical concepts and the behavior of the experiment. Several OLEs do not have specific goals for the student to attain. If there is not a defined experimental goal that the student is required to attain, it is not easy to infer the current knowledge state in an OLE. In addition, students’ learning processes are not yet completely understood. It is expected that the student will be capable of constructing knowledge while interacting with the system. As a result, specific objectives are needed in order to attain and enable an effective assessment of the learning goals (Noguez J. & Sucar E., 2005). Even so, it is hard to detect and determine how much does the student actually knows at a given point in time, what they do not know and what skills have or have not been being acquired by the student, based only on the student’s interaction with the OLE. Given the inherent uncertainty of these tasks, a decision model capable of reasoning under ambiguity is required.
To face the aforementioned learning problems, our aim is to include elements from Intelligent Tutoring Systems (ITSs) inside ALSs, which are capable of giving students (or users in general) the benefits of an adaptive learning environment. An ALS takes care of the balance between free exploration capabilities and the tutoring labor based on the best pedagogical action. For example, decisions such as when to interrupt a given experiment and to provide on time help for the user, and actions such as following-up the student’s performance and planning the forthcoming task.
To handle uncertainty, student models, based on Bayesian Networks (BNs), have already been developed for ITSs (Pearl, 1988). These BNs are useful for diagnosing – or inferring – the student’s current cognitive state, using observable data (VanLehn, 2001; Murray, 200; Mayo, 2001). However, the effort required to build the network structure, to obtain and define the related parameters, and to manage the computational complexity of the inference algorithms, makes the application of these types of models very difficult. This is particularly true in real time situations, such as simulators (Adams, 2007). Finding a general model for a variety of experiments, scenarios and domains is an additional challenge. In terms of development, efforts to create tutors for all simulators types are very laborious and time consuming. We have proposed using Probabilistic Relational Models (PRMs), which provide a new approach to student/user modeling, integrating the expressive power of BNs with the advantages of relational models (Noguez J. & Sucar E., 2005). PRMs enable starting with the domain problem, like object-oriented modeling, and continuing by defining the elements of probabilistic modeling. A general structure for each module of the ITS was designed. The first step was to identify the classes in the model. The next step was to define the dependency model at class level, allowing it to be used for any object in the class, thus facilitating the understanding of the model. The classes and their relationships provide a general schema. Based on this schema, a skeleton is derived that integrates the relevant variables and their dependencies into BNs or Dynamic Decision Networks (DDNs). Finally, a particular BN or DDN is obtained from the skeleton for each specific experiment in certain domains.