A Machine Learning Approach to Leverage Individual Keyboard and Mouse Interaction Behavior From Multiple Users in Real-World Learning Scenarios

Abstract

There is strong evidence that emotions influence the learning process. For this reason, we explore the relevance of individual and general mouse and keyboard interaction patterns in real-world settings while learners perform free text tasks. To this end, we have modeled users' mouse movements and keystroke dynamics with data mining techniques, building on previous related research and extending it in terms of some critical modeling issues that may have an impact on detection results. Inspired by practice in affective computing where physiological sensors are used, we argue for the creation of an interaction baseline model, as a reference point in the way how learners interact with the keyboard and mouse. To make the proposed affective model feasible, we have adopted a simplified 2-D self-labeling approach for labeling the users' affective state. Our approach to affect detection improves results when there is a small amount of data instances available and does not require additional affect-oriented tasks from the learners. Specifically, learners are only asked to self-reflect their emotional state after finishing the tasks and immediately selecting two values in the affect scale. The approach we have followed aims to distill two types of interaction patterns: 1) within-subject patterns (from a single participant) and 2) between-subject patterns (across all participants). Doing this, we aim to combine both the approaches as modeling factors, thus taking advantage of individual and general interaction patterns to predict affect.

Keywords

• Affective computing
• data mining
• keystrokes dynamics
• learner modeling
• learning analytics
• MOKEETO