Deep Collaborative Filtering for Enhanced Learning Outcome Modeling

  • Author / Creator
    Chen, Fu
  • The digital learning and assessment movement has contributed to an explosion of structured and unstructured learner data (e.g., learner problem-solving product and process data). This calls for new developments in large-scale learning outcome modeling to optimally address the variety, volume, uncertainty, and velocity of big data in education. Existing learning outcome modeling techniques, such as psychometric measurement models and Bayesian models, typically require structured product data and fail to account for process data. Moreover, most of them are incapable of learning associations between items and latent skills. Leveraging the advantages of collaborative filtering (CF) used in recommender systems, this study proposes three novel deep learning-based CF approaches — SDCF, LogCF, and LogSDCF — to model both product and process data for enhanced learning outcome modeling. The three models are also capable of discovering item-skill associations from the data without expert information. Specifically, SDCF is developed to model product data sequentially by predicting learners’ next item responses based on their history of item responses; LogCF is proposed to model both product and process data to predict learners’ missing or future item responses when item responses are not in a sequential form; LogSDCF is devised to model both product and process data to predict learners’ future item responses based on their response history when items are presented in a sequential form. To evaluate the effectiveness of the proposed approaches, the three models were compared with conventional learning outcome modeling approaches using both simulated and real-world datasets. Results showed that all three approaches achieved higher prediction accuracy of learners’ future or missing item responses than their baselines. Moreover, the proposed approaches were found to be promising in discovering item-skill associations without expert input.

  • Subjects / Keywords
  • Graduation date
    Fall 2021
  • Type of Item
  • Degree
    Doctor of Philosophy
  • DOI
  • License
    This thesis is made available by the University of Alberta Libraries with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.