An emotion analysis in learning environment based on theme-specified drawing by convolutional neural network.

Area of Science:

• Educational Psychology
• Artificial Intelligence
• Computer Vision

Background:

• Learner emotions significantly impact cognitive processes like attention and memory.
• Manual analysis of emotions through hand-drawn paintings is subjective and inefficient for clinical practice.
• There is a need for objective and efficient methods to assess emotional status in learning environments.

Purpose of the Study:

• To explore the effectiveness of a convolutional neural network (CNN) model for analyzing learner emotions from hand-drawn paintings.
• To address the subjectivity and inefficiency of manual, psychologist-led emotional analysis.
• To develop an automated system for reflecting learner emotional status in educational contexts.

Main Methods:

• A CNN model was developed to analyze 100 × 100 pixel painting images.
• Learner emotional status was self-reported via questionnaire and validated by a psychologist, serving as training labels.
• The model underwent convolutional, full-connected, and classification operations to learn image features and classify emotions.

Main Results:

• The CNN model achieved an accuracy of 72.1% in classifying learner emotional status from paintings.
• The model demonstrated effectiveness in learning features from pixel data to high-level semantic mappings.
• Experimental validation involved 2,103 learners, with 2,000 paintings used for training and testing.

Conclusions:

• The developed CNN model provides an effective, objective, and efficient approach to emotional analysis in learning environments.
• Automated painting-based emotional analysis using AI can overcome the limitations of traditional subjective methods.
• This technology holds potential for improved monitoring and support of learner well-being and academic performance.