Computer-based intelligent support for moderately Ill-structured problems

Abstract

Problems are often categorized into two types: ill-structured and well-structured, in the context of education/learning, cognitive science and artificial intelligence. Then, from an educational viewpoint, ill-structured problems are further more important because they are useful to promote a learner to think about learning target deeply and to master computational or logical thinking skills, including metacognition. This paper proposes an additional characterization of problems by using two factors, (1) well/ill-structured domain model and (2) well/ill-structured problem setting. Based on this characterization, “moderately ill-structured problems” are defined as a category of problems specified by “well-structured domain model” and “ill-structured problem setting”. If a problem is set in well-structured domain model, it is possible to realize computer-based monitoring and diagnosis of learner's activities for the problem. If the problem setting is ill-structured, for example, open-ended, a learner is required to engage in the problem as ill-structured one. Therefore, moderately ill-structured problems are promising to realize computer-based intelligent support for solving ill-structured problems, while keeping educational advantages of ill-structured problems. In this paper, a definition of moderately ill-structured problems is described. Then, using the arithmetic word problems as an example of learning target domains, this paper describes (1) well-structured domain model of arithmetic word problems, and (2) design of moderately ill-structured problem as “problem-posing assignment” based on the domain model. Moreover, (3) implementation of an intelligent learning environment that requests a learner to solve ill-structured problems as problem-posing is introduced. The environment has functions to diagnose learner's behaviors and to give individual feedback.