Emotional Variance Analysis: A new sentiment analysis feature set for Artificial Intelligence and Machine Learning...

Area of Science:

• Computational Linguistics
• Educational Data Mining
• Affective Computing

Background:

• Sentiment Analysis (SA) extracts affective states from text, with applications in reviews and mental state assessment.
• Existing SA techniques may not fully capture nuanced emotional dynamics crucial for performance prediction.

Purpose of the Study:

• Introduce Emotional Variance Analysis (EVA) as a novel SA feature set.
• Evaluate EVA's efficacy in profiling sentiment variations and predicting academic performance in an educational context.

Main Methods:

• Developed and applied the Emotional Variance Analysis (EVA) feature set to student journals from an Experiential Learning (EL) course.
• Utilized a Multi-Layer Perceptron (MLP) machine learning model for performance prediction.
• Compared EVA's predictive accuracy against traditional Natural Language Processing (NLP) and SentimentR features.

Main Results:

• EVA achieved an 88.7% overall accuracy in predicting student experiential learning grades.
• EVA significantly outperformed NLP (76.0%) and SentimentR (58.0%) features, with improvements of 15.8% and 51.7%, respectively.
• Demonstrated EVA's capability in profiling variations in sentiment polarity and intensity.

Conclusions:

• Emotional Variance Analysis (EVA) is a robust feature set for sentiment analysis, particularly effective in educational data mining.
• EVA's ability to capture emotional instability offers potential applications in mental health profiling and consumer behavior analysis.
• The developed EVA feature set is compatible with various Artificial Intelligence (AI) and Machine Learning (ML) applications.