Global User-Level Perception of COVID-19 Contact Tracing Applications: Data-Driven Approach Using Natural Language...

Area of Science:

• Artificial Intelligence
• Natural Language Processing
• Public Health Technology

Background:

• COVID-19 contact tracing apps are widely used, but their effectiveness and user reception require continuous monitoring.
• Existing methods for analyzing user feedback are manual and time-consuming.
• There is a need for automated solutions to rapidly assess user sentiment towards these critical health applications.

Purpose of the Study:

• To evaluate the efficacy of AI and NLP techniques for automated sentiment analysis of COVID-19 contact tracing app reviews.
• To develop and validate a sentiment analysis framework for classifying user review polarity.
• To create a large-scale annotated dataset for advancing research in this domain.

Main Methods:

• A pipeline involving manual annotation of user reviews through crowdsourcing.
• Development and training of AI models, including classical and deep learning approaches, for sentiment classification.
• Collection and annotation of a substantial dataset of user reviews for COVID-19 contact tracing applications.

Main Results:

• Achieved high performance with an average F1 score of 94.8% across 8 methods and 3 tasks, demonstrating the feasibility of the proposed solution.
• Generated a large-scale benchmark dataset comprising 34,534 manually annotated user reviews.
• Successfully developed a proof-of-concept web application for rapid sentiment analysis.

Conclusions:

• AI and NLP techniques effectively analyze and classify user sentiments in app reviews, outperforming manual methods in speed and accuracy.
• The developed framework and benchmark dataset can serve as a rapid surveillance tool for monitoring mobile applications during emergencies.
• Automated sentiment analysis enables quicker identification of app issues and facilitates timely improvements without intensive user design processes.