Mining Behavioral Patterns from Response Time: A Sequential Modeling Approach

Response times (RTs) in large-scale educational assessments are increasingly used to identify disengaged response behaviors such as rapid guessing, idling, cheating, or item pre-knowledge. While these applications provide valuable insights, they typically treat each RT in isolation, overlooking the sequential dependence among them. For example, a student who spends excessive time on early items is more likely to engage in rapid guessing later or fail to complete the test on time. To capture these dynamics, RTs can be conceptualized and modeled as sequential data. In this talk, I will demonstrate how response time sequences can be leveraged using advanced methods inspired by deep learning and natural language processing (e.g., n-grams). Two case studies will be presented: (1) predicting early quitting behavior in the e-TIMSS Problem Solving and Inquiry tasks, and (2) predicting students’ test scores in a computerized adaptive test. I will conclude by discussing the implications of modeling RT sequences for researchers and practitioners interested in better understanding and predicting learner behaviors in digital assessments.

Bio
Okan Bulut is a Professor in the Measurement, Evaluation, and Data Science program and a researcher at the Centre for Research in Applied Measurement and Evaluation at the University of Alberta, Canada. He teaches courses on computational psychometrics, machine learning, and statistical modeling. He also regularly offers workshops and online courses on data mining, big data modeling, data visualization, and statistical analysis using R, SAS, and Python. His research focuses on the intersection of artificial intelligence (AI), educational data mining, and learning analytics. In particular, he applies AI-driven algorithms and natural language processing to extract insights from complex educational data and to develop personalized assessment and learning tools.   
http://www.okanbulut.com