Deep Learning Model for COVID-19 Sentiment Analysis on Twitter

Area of Science:

• Computational Social Science
• Natural Language Processing
• Public Health Informatics

Background:

• Social media platforms like Twitter offer valuable insights into public sentiment during health crises.
• The COVID-19 pandemic significantly influenced public mood, necessitating methods to gauge population-level emotional responses.
• Mexico experienced severe COVID-19 waves, making sentiment analysis crucial for understanding public perception.

Purpose of the Study:

• To analyze the sentiment of the Mexican population on Twitter during a major COVID-19 wave.
• To develop and evaluate a high-precision sentiment analysis model for Spanish-language COVID-19 discussions.
• To compare the performance of a dedicated Spanish Transformer model against multilingual models and other classifiers.

Main Methods:

• A mixed, semi-supervised approach utilizing lexical-based data labeling.
• Training of two Spanish-language Transformer models specifically for COVID-19 sentiment analysis.
• Comparative analysis involving ten multilingual Transformer models and traditional classifiers (SVM, Naive Bayes, Logistic Regression, Decision Trees).

Main Results:

• The dedicated Spanish-language Transformer model demonstrated superior precision in sentiment analysis compared to multilingual and traditional models.
• The developed model effectively captured public sentiment regarding COVID-19 on the Mexican Twitter network.
• Performance evaluation confirmed the efficacy of specialized language models for domain-specific sentiment analysis.

Conclusions:

• A specialized Spanish Transformer model is highly effective for analyzing COVID-19 related public sentiment on social media.
• This approach provides a valuable tool for monitoring public opinion during health emergencies in Spanish-speaking regions.
• The findings underscore the importance of language-specific models in Natural Language Processing tasks for accurate social science research.