Sentiment Analysis on COVID-19 Twitter Data Streams Using Deep Belief Neural Networks.

Area of Science:

• Computational Social Science
• Public Health Informatics
• Artificial Intelligence

Background:

• Social media platforms like Twitter facilitate rapid information dissemination, influencing public opinion during health crises.
• The COVID-19 pandemic highlighted the dual threat of misinformation and the value of public sentiment analysis for health policy.
• Understanding public attitudes towards measures like social distancing is vital for effective public health interventions.

Purpose of the Study:

• To analyze public sentiment on social distancing measures during the COVID-19 pandemic using Twitter data.
• To develop and evaluate an automated sentiment analysis model for classifying public opinion from tweets.
• To provide insights for public health officials and decision-makers based on sentiment analysis.

Main Methods:

• Employed text preprocessing techniques including tokenization, filtering, stemming, and N-gram model construction (specifically bigram).
• Utilized a Deep Belief Neural Network (DBN) architecture for tweet classification.
• Integrated a pseudo-labeling strategy to enhance DBN performance and convergence speed, preserving computational efficiency.

Main Results:

• The proposed sentiment analysis model achieved a classification accuracy of 90.3% using the DBN classifier with bigram N-grams.
• Pseudo-labeling improved classification performance and keyword extraction precision.
• The model demonstrated effectiveness in capturing public sentiment regarding social distancing.

Conclusions:

• The developed DBN model with pseudo-labeling and N-gram analysis provides a highly accurate method for social media sentiment analysis.
• This approach can effectively gauge public opinion on critical health issues like social distancing during pandemics.
• Findings can assist public health professionals in tailoring interventions based on location-specific public sentiment.