State-Aware Deep Item Response Theory using student facial features.

Area of Science:

• Educational Psychology
• Artificial Intelligence
• Psychometrics

Background:

• Item Response Theory (IRT) traditionally models student ability and item difficulty.
• Student cognitive and affective states during testing can influence response behavior.
• Facial expressions provide non-verbal cues to these internal states.

Purpose of the Study:

• To develop a novel approach integrating deep learning and IRT to analyze facial expressions for improved response prediction.
• To introduce the State-Aware Deep Item Response Theory (SAD-IRT) model with a new student state parameter.
• To evaluate SAD-IRT's effectiveness in predicting student responses and understanding their states.

Main Methods:

• Utilized deep learning techniques to extract facial features from student test-taking videos.
• Developed the SAD-IRT model incorporating a latent student state parameter regressed from facial features.
• Compared SAD-IRT's predictive performance against standard IRT and deep learning IRT models.

Main Results:

• SAD-IRT significantly improved prediction performance of student responses compared to baseline models.
• The model maintained accurate estimations of student ability and item difficulty parameters.
• SAD-IRT demonstrated early prediction capabilities, forecasting response outcomes before completion.

Conclusions:

• Facial expression analysis via deep learning can enhance IRT models.
• The student state parameter offers insights into subjective test item difficulty.
• SAD-IRT paves the way for more personalized and responsive educational assessments.