Improving user modeling via the integration of learner characteristics and learner behaviors

• Main Author: Dalkir, Kimiz L
• Format: Others
• Published: 1997
• Online Access: http://spectrum.library.concordia.ca/216/1/NQ25914.pdf; Dalkir, Kimiz L <http://spectrum.library.concordia.ca/view/creators/Dalkir=3AKimiz_L=3A=3A.html> (1997) Improving user modeling via the integration of learner characteristics and learner behaviors. PhD thesis, Concordia University.

Three major disciplines: educational psychology, cognitive science and artificial intelligence, were critically surveyed to identify useful variables for learner modeling in order to identify the subset of variables that proved to be useful in modeling individual learners as they interacted with a computer-based learning environment. The research study first critically assessed the contextual validity and usefulness of a priori measures to provide initial or default values for a stereotypical learner model. These measures included a pretest, a questionnaire and two learning style inventories, the Kolb LSI and the Entwistle ASI. In addition, the utility of Artificial Neural Networks (ANNs) was assessed to establish whether they provide supplementary or complementary information for the objective of creating an adaptive learner model. Differences in interaction patterns with the learning environment were analyzed using ANNs and statistical analyses, to identify on-line learner behavioral variables that were valid and useful in updating the stereotype learner model. The instructional validity of the learning environment was established as students were found to have spent time interacting with the system, they attended to the material presented and they were found to have learned the content. Significant learning was found, as assessed by pretest-posttest differences. Of the Educational Psychology variables, only the Entwistle ASI proved to be useful as an a priori measure in this context. Students with high scores on both the reproducing and meaning orientation dimensions performed better on the posttest. In addition, expected learner profiles, as extrapolated from the ASI, actually occurred as students interacted with the system. Finally, Artificial Intelligence approaches, in the form of Artificial Neural Networks (ANNs) were superimposed on the a priori categorization established by the learning style categories. Conventional statistical cluster analysis and ANN pattern recognition on learner trace data produced as students interacted with the learning materials both produced very similar classifications of students. It thus appears to be possible to obtain, effectively, the same data from an ongoing dynamic assessment of learners as it is from a priori measures, rendering the latter redundant in this context. Thus the use of ANNs can prove useful as a dynamic data gathering and analysis system in real time to make instructional adjustments and recommendations. The potential advantage of dynamic models over a priori measures is that they continue to evolve as learner needs change, continually updating the learner model and thus enabling the learner model to keep pace with an instructional system endowed with adaptive capabilities. Future research could build on the exploratory data generated by this study, examining both the variables which may inform the creation of an adaptive interface, as well as using the ANN-based methodology created here