Modeling Student Teachers' Information-Seeking Behaviors: Implications for Adaptive Scaffolding of Lesson Planning

• In Event: Meaningful Scaffolds to Foster Learning in Interactive Learning Settings

Abstract

Objective. An important challenge for the design of computers as metacognitive tools is to provide a mechanism for detecting when and where adaptive scaffolding of self-regulated learning is needed. This study demonstrates the usefulness of an educational data mining technique to discover learner patterns that are indicative of successful information seeking, a prerequisite to effective learning. The proposed method has direct applications for the design of a recommender system that determines when and where student teachers require scaffolds to help them find useful and relevant online resources while designing a lesson plan.

Rationale. nBrowser is an intelligent web browser designed to scaffold student teachers’ self-regulated learning while navigating the web to translate information about educational technologies into technology-rich learning experiences. The learner model implemented in the system allows nBrowser to continually updated a computational representation of online resources based on learner behaviors, including their ratings for the usefulness of these resources towards task performance (Poitras & Fazeli, 2016). However, teachers seldom report ratings while working on their lesson plans. As such, we posit that the mouse cursor positions and movements logged by the system could serve as an indicator of the utility of information found by the students.

Methods. Student teachers were randomly assigned to either a dynamic or static version of nBrowser. The dynamic version of nBrowser used student ratings of the usefulness of online resources to inform the recommendations, whereas the static version did not use student ratings to update the model. The log trace data was aggregated at the level of page load events in which a rating was submitted by student teachers, which was dichotomized into a positive and negative rating category.

Results. A decision tree model based on time-series features that characterized mouse cursor position and movements was trained to predict the student teachers’ usefulness ratings and allow nBrowser to generate real-time suggestions. The features consist of the pixel coordinates of the mouse cursor (i.e., position) and the distance traveled from one point to another (i.e., movement) within a 10 second window (i.e., moving in 1 second increment; 2 second latency between feature extraction). The best candidate model attained 62.62% predictive accuracy of positive and negative ratings for the usefulness of online resources.

Significance. The underlying assumption is that the learner behavior detection model can reduce the amount of time and resources necessary for nBrowser to adjust system recommendations of online resources. We discuss the implications of the learner behavior detect model for the delivery of adaptive scaffolds through the recommender system.



References

Poitras, E., & Fazeli, N. (2016). Mining the edublogosphere to enhance teacher professional development. In Shalin Hai-Jew (Ed.), Social Media Data Extraction and Content Analysis. IGI Global.

Authors

• Eric G. Poitras, University of Utah

• Tenzin Doleck, McGill University

• Susanne P. Lajoie, McGill University