Deep Learning-Based Methods for Sentiment Analysis on Nepali COVID-19-Related Tweets

Area of Science:

• Computational Social Science
• Natural Language Processing
• Mental Health Research

Background:

• The COVID-19 pandemic has significantly impacted global mental health, with psychological distress often exacerbated by social and informational environments.
• Understanding public sentiment and psychological states during health crises is crucial for effective public health interventions.
• Online social media platforms, like Twitter, offer a rich source of real-time data reflecting public sentiment and concerns.

Purpose of the Study:

• To analyze public sentiment and psychological states related to COVID-19 in Nepal using Twitter data.
• To develop and evaluate novel feature extraction methods and Convolutional Neural Network (CNN) models for sentiment classification of Nepali tweets.
• To introduce a new benchmark dataset, NepCOV19Tweets, for COVID-19 sentiment analysis in the Nepali language.

Main Methods:

• Proposed three distinct feature extraction techniques: fastText-based (ft), domain-specific (ds), and domain-agnostic (da).
• Developed three Convolutional Neural Network (CNN) models tailored to the proposed feature representations.
• Ensembled the three CNN models into a unified, end-to-end system for sentiment classification.
• Created and utilized the NepCOV19Tweets dataset, comprising 3 classes (positive, neutral, negative) of Nepali tweets related to COVID-19.

Main Results:

• The proposed feature extraction methods demonstrated strong discriminating capabilities for sentiment classification.
• The developed CNN models exhibited robust and stable performance when applied to the extracted features.
• Experimental results validated the effectiveness of the ensemble CNN approach for sentiment analysis.
• The NepCOV19Tweets dataset proved effective for evaluating the proposed methods.

Conclusions:

• The study successfully developed and validated novel methods for COVID-19 sentiment analysis in Nepali Twitter data.
• The proposed feature extraction and CNN models are effective for classifying sentiment, offering insights into public psychology during the pandemic.
• The NepCOV19Tweets dataset serves as a valuable resource for future research in Nepali-language sentiment analysis and computational social science.