Instructional planning in intelligent evolutive tutoring systems from a case-based reasoning approach
Jon Ander Elorriaga · AI Communications · 1999
The field of Artificial Intelligence in Education (AI-Ed) uses Artificial Intelligence techniques with the aim of constructing intelligent tutoring systems. The ‘intelligence’ of these systems is determined by their degree of adaptation to the learning characteristics and behaviour of the students [4]. Thus, the ability to adapt the instruction to the student is based on two issues: the learner model and the instructional planning. The research on the second aspect has been centered mainly on the dynamic aspect of instructional planning while few researchers have tried to add learning capabilities to instructional planners. The goal of this project is to study the viability of an instructional planner that is able to gather the results of its previous experiences and construct plans, taking this information into account. Concretely, the instructional planner uses case-based reasoning as a consequence of its inherent learning capability. Moreover, the planner uses the opinion of both the learner and the teacher in the instructional decision. Instructional planning is a main task in automatic intelligent tutoring; it consists of determining the suitable sequence of instructional goals and actions to promote learning at the same time that it provides consistency, coherence and continuity throughout an instructional session [3]. The debate about the necessity and advisability of an instructional plan, which guides the development of the learning session has been active these last years. In our opinion, instructional planning together with student modelling are the keys for individualised instruction. Through instructional planning, intelligent tutoring systems (ITSs) use pedagogical knowledge to select and order the most appropriate topics and learning activities for each concrete student. On the other hand, instructional planning does not mean necessarily control, but a guide to develop the session in the most appropriate way for the student. Case-based reasoning (CBR) is a relatively new paradigm for problem solving and learning that simulates basic human reasoning: when people have to solve a new problem, they recall similar situations and apply the same solution to the current problem [1]. CBR exhibits an inherent learning capability because it stores each experience and uses these in the future. The core of this thesis is a case-based instructional planner (CBIP). CBIP is composed of a knowledge structure (Instructional Plan Memory) and three functional modules (Generation Component, Learning Component and Assessment Component). The Instructional Plan Memory stores the experiences of the system; its organisation is based on Schank’s Dynamic Memory Model [2]. The process of the planner is shared out between the Generation and Learning Components. The Generation Component is responsible for retrieving similar instructional experiences from the memory and adapting them to the characteristics of the current situation. The Learning Component is responsible for the evaluation of the results produced by the instructional plans and for updating the memory accordingly. This module performs learning by storing the new experiences with their results. Both modules use the Assessment Component, which contains a set of heuristic formulae that evaluate different objects involved in the instructional planning process. The Hybrid Self-Improving Instructional Planner (HSIIP) is based on the CBIP. This approach for en-