Mood detection and prediction using conventional machine learning techniques on COVID19 data.

Area of Science:

• Computational linguistics
• Affective computing
• Machine learning

Background:

• Emotion detection from text is crucial for fields like psychology and marketing.
• Advancements include sophisticated learning approaches beyond simple frequency analysis.
• Accurate prediction of real-time streaming text moods is a key objective.

Purpose of the Study:

• To compare the performance of nine conventional learning methods for emotion detection.
• To analyze mood variance over time using a wide array of 25 moods.
• To evaluate classifiers for predicting multiple emotions in streaming text.

Main Methods:

• Extensive comparison of nine conventional learning methods using precision, recall, F1, and accuracy metrics.
• Development of an Android application (Citizens' Sense) for text collection and analysis.
• Testing classifier performance on Twitter data related to COVID-19.

Main Results:

• Random Forest, Decision Tree, and Complement Naive Bayes classifiers showed marginally superior performance.
• These conventional classifiers offer near real-time predictions and flexibility in feature engineering.
• Analysis confirmed the variance of mood over time and supported classifier findings.

Conclusions:

• Conventional machine learning classifiers are effective for real-time emotion detection and mood analysis.
• Random Forest, Decision Tree, and Complement Naive Bayes are recommended for streaming text emotion classification.
• The study highlights the utility of these methods for understanding public sentiment and mood dynamics.