Better Calibration of Learner Knowledge Tracing Using Contextualized Information (Poster 2)

• In Event: AERA Virtual Poster Session 7
  In Poster Session: Poster Session

Abstract

Several studies were conducted to explore learner models using log data generated by intelligent tutoring systems. However, most of these studies focus on modeling student response data to track students’ learning status and predict their future performance. The present study implements DKT-CI, a novel Deep Knowledge Tracing (DKT) model that incorporates contextualized information such as response time and number of hints for learner knowledge tracing. We compared the prediction accuracy of DKT-CI to that of three baseline models: IRT, BKT, and DKT that only model response data. Results show that the DKT-CI outperforms the baseline models on estimating student probabilities of skill mastery. Findings of the present study demonstrate the potential of deep learning approaches in modeling process data.

Authors

• Chang Lu, Shanghai Jiao Tong University

• Maria Cutumisu, McGill University