CSE Assistant Professor Tempestt Neal and her PhD student Meghna Chaudhary (pictured right) have recently co-authored a paper titled, “On the use of aspect-based sentiment analysis of Twitter data to explore the experiences of African Americans during COVID-19,” published in Nature Scientific Reports. Their work explores the application of Natural Language Processing on Twitter data to investigate the experiences of African Americans during the COVID-19 pandemic. The research was inspired by the disparities underscored by the U.S. Center for Disease Control and Prevention, notably the disproportionate impact of the coronavirus on the African American population, resulting in higher death rates compared to other demographic groups.
Recognizing the urgency to understand the experiences, behaviors, and opinions of African Americans during the COVID-19 pandemic, the study utilized Twitter data for an aspect-based sentiment analysis, enabled by a multimodal pipeline to filter relevant tweets. The study was funded by USF’s COVID-19 Rapid Response Research Program and Microsoft AI for Health. Alongside Professor Neal advising this work include Co-PIs Dr. Kristin Kosyluk, an Assistant Professor in the Department of Mental Health Law and Policy and director of the STigma Action Research (STAR) Lab, and Dr. Sylvia Thomas, Professor in Electrical Engineering and director of the Advanced Membrane and Materials Bio and Integration Research (AMBIR) Lab.
“The genesis of this study can be traced back to a thought-provoking discussion with my advisor, Professor Tempestt Neal, about the disproportionate impact of the COVID-19 pandemic on African American communities,” said Meghna Chaudary. Meghna is a PhD student in the Department of Computer Science and Engineering, specializing in Natural Language Processing, targeted sentiment analysis, and pattern recognition. This builds on her master’s work in India, which focused on gauging student sentiments towards university events.
“My role was to dissect text data for patterns in the emotional responses of African Americans affected by the pandemic. The objective was to discern the specific sentiments and their corresponding entities over time within this demographic,” said Meghna.
The ambitious project required processing a vast corpus of Twitter data--approximately a billion tweets. Meghna engineered a machine learning pipeline to deduce user race based on Twitter profile images and language cues. This narrowed the scope of the analysis to around four million relevant tweets. The team then analyzed the results using advanced linguistic models which detect the opinions of an author via implicit entities in the text.
This form of sentiment analysis identifies both emotional tones and the topics discussed. Intriguing semantic relationships cropped up between words, such as “covid” and “exhausted.” Topics, such as food insecurity and vaccine hesitation, were also prevalent. The study also explored the evolution of word usage throughout the year. For instance, the word “masks” frequently surfaced in early 2020, and the tone reflected the debates surrounding mask usage. Although, by mid-year, the sentiment shifted to a critical stance on the public’s reluctance. Overall, the study showed a predominantly negative tone, revealing how the pandemic impacted the narratives of African American Twitter users.
“Building on my background in stylometry, my lab has undertaken various projects in natural language processing. One such significant endeavor is the project under discussion.” said Professor Neal. Her expertise lies in biometric recognition and stylometry, both focusing on identifying individuals or verifying their identity claims. Professor Neal’s previous work involved utilizing mobile device sensing data and text data for these purposes.
“Our research found a fitting platform in Nature’s Scientific Reports,” said Professor Neal. Nature Scientific Reports is the 5th most-cited journal globally, covering natural sciences, psychology, medicine, and engineering. Nature.com reaches more than 8 million visitors each month.
“I am thrilled about the publication of our work in the magazine, as it provides a platform to disseminate our findings to a broader audience. It is my belief that making this information accessible will not only enlighten the public but also encourage further scholarly inquiry into societal impacts like the ones we’ve examined,” said Meghna.