Adaptive training based upon computerized knowledge assessment

One of the most praised qualities of computer assisted training (CAT) is its ability to deliver individualized training. In CAT, the pace of individual learning is not determined by the learning group’s average, or maximum, or minimum learning pace. It adapts to the individual’s capabilities. Even more, the content can be tailored to the individual’s background knowledge and personal training needs. Current state-of-the art CAT systems support individualization of the training process by allowing free or flexible navigation within the training content, as opposed to a rigid linear content presentation controlled by the computer. However, this flexibility imposes an additional burden on both the instructional designer and the learner. For example, the instructor must select the appropriate sequence of exercises according to the successes and failures of the learner. Or, the learner must explore the whole training content and decide what is appropriate. However, dynamic modeling of a user’s current knowledge state can alleviate this problem. With such capabilities, it becomes possible to specify the learning material’s prerequisites and appropriate level of expertise in such a way as to avoid presenting too advanced or trivial content to a given user. Moreover learning material can be tailored in a very specific manner, targeting only the necessary content for a given learner’s knowledge state. We present a method for the dynamic modeling of a user’s knowledge state. The method induces, from empircal data, the learning order of knowledge units (KU—they can represent mastery of concepts or skills). This ordering is thereafter used to infer an individual’s knowledge state (see Falmagne et al., 1990, for the cognitive foundations of this method). Through a sampling of a person’s knowledge state, the system assesses the probability that this person knows any given KU.