Abstract:

The precise determination of student learning has always been an important point for several purposes such as the final student evaluation, the selection of remedial tactics for correcting learning errors and for the efficiency of learning systems. The automatic obtainment of errors has usually relied on the difference between the student’s and the human expert behaviour, which is not always apparent.

The difficulty starts when the system tries to assess that difference of behaviour, mostly in the case of complex behaviours. The paper presents new methods developed not only for error determination but also for error classification between shallow and deep errors and for obtaining the most probable reasons of the error, providing that way the most suitable remedial tactics.
The methods are assembled into a cycle with the phases: error determination, initial evaluation of the error, shallow analysis, deep analysis and final evaluation, all of them carried out by means of simple graph operations, mostly by comparing the human expert model with the student’s model. The methods can be applied automatically provided the learning system can manage the student learning model (usually a genetic graph) and the human expert model or the learning domain model.
Some results of the comparison between shallow and deep analysis carried out automatically by the system and a similar analysis made by instructors using traditional methods, are given in the paper.

The results of these methods can be used to design automatically remedial tactics especially suitable for the obtained errors. Several examples of this are included.

Keywords:

Learning assessment, error detection, shallow error, deep error, student evaluation.