Efficacy of personalized models in discriminating high cognitive demand conditions using text-based interactions

Area of Science:

• Human-Computer Interaction
• Cognitive Science
• Psychology

Background:

• High cognitive demand can negatively impact psychological, physical, and behavioral functions.
• Existing methods for measuring cognitive demand are impractical for continuous monitoring and lack personalization.
• Individual differences in cognitive demand response are not adequately addressed by current approaches.

Purpose of the Study:

• To develop and validate individualized models for monitoring cognitive demand using keystroke and linguistic text features.
• To differentiate between high and low cognitive demand conditions.
• To inform the design of personalized health management systems capable of continuous cognitive demand monitoring.

Main Methods:

• Constructed within-subject and between-subjects models using keystroke dynamics and linguistic markers from typed text.
• Evaluated model performance using Area Under the Curve (AUC) metrics.
• Identified common features indicative of cognitive demand across participants.

Main Results:

• Personalized models demonstrated significantly higher accuracy (Mean AUC=0.826) compared to a generic model (AUC=0.714) and chance.
• AUCs for individual models ranged from 0.679 to 0.953, indicating substantial individual variability.
• Key features included negative emotion rate, lexical diversity, long word frequency, and word count.

Conclusions:

• Significant individual differences exist in cognitive demand response.
• Future cognitive demand monitoring systems should incorporate adaptive, individualized features.
• This research offers a novel combination of text and keystroke analysis for detecting high cognitive demand in Human-Computer Interaction contexts.