Person-Oriented Learner Modeling Approaches to Assessing, Predicting, and Tracking Learning in Technology-Rich Environments

• In Event: 41.025 - Learning, Engagement, and Assessment in Technology-Rich Learning Environments

Abstract

Objectives: The purpose of the present study was to determine the effectiveness of person-oriented learner modeling approaches in the context of two different Technology-rich learning environments (TRE), BioWorld (Lajoie, 2009) and Metatutor (Azevedo, 2009). TREs provide learners with unprecedented experiences that allow for dynamic interactions between environmental variables and learner variables, which in turn lead to meaningful learning and multidimensional growth. Traditional off-line measures and statistical modeling approaches are limited in modeling the learner based on such dynamical interactions while adaptively scaffolding learning progressions. A representation of the learner model needs to be constantly updated by the TRE system in order to determine the type of mediation (corrective feedback, indirect hints), the degree of learner control, and timing of mediation and further to adapt subsequent tasks to the current state of the learner. The success of adaptive learner modeling requires theoretically sound, empirically justifiable, and pedagogically meaningful approaches to modeling the learner. Such adaptive learner modeling involves multiple steps including: 1) gathering information about the learner as he or she interacts with the TRE; 2) developing a learner model that captures the current state of the learner in multidimensional domains; 3) adaptively selecting tasks and mediation types, timing, and content; and 4) presenting highly individualized content and feedback using appropriate modes for maximal utilization (Jonassen, 2004).
Methods and Results: The present study utilized multi-modal, multi-task, and mixed methods data sets (computer logs, self reported questionnaires, quizzes) for the modeling of the learner in each of the studied TREs. For each TRE platform, different sets of learner variables were identified for classifying learners according to distinct patterns of learner behaviours, predicting learning outcomes, and tracking growth over tasks through person-oriented measurement approaches. Both qualitative and quantitative learner-modeling approaches were employed in order to probablistically estimate the current state of the learner, conceptualize the relationships among the learner variables, and graphically represent heterogeneous individual learner models resulting from complex interactions among intrapersonal variables. Specifically, we applied various person-oriented measurement approaches including latent class modeling, latent profiling modeling, latent trait modeling, and exploratory hierarchical classification methods (Collins & Lanza, 2010; Lazarsfeld & Henry, 1968). The learner variables identified through integrative mixed methods data analyses included both cognitive and noncognitive learner traits as well as learner behavioral frequency-based variables from computer trace data.
Significance: The study results show the effectiveness of learner modeling approaches in terms of classification and prediction accuracy. Individualized graphic representations of learner models are shown to provide diagnostic information about the current state of the learner as well as the ‘multiple zones of proximal development.’ We discuss how such learner representations can be used for adaptive scaffolding for renewal of the learner model. We further discuss the potential of person-oriented learner modeling approaches for designing and updating the system across different TRE platforms.

Authors

• Eunice Eunhee Jang, University of Toronto

• Clarissa Lau, University of Toronto

• Jeanne Sinclair, University of Toronto