This project is maintained by edgeslab
University of Illinois at Chicago
PIs: Elena Zheleva, Barbara Di Eugenio, Liz Marai, Andrew Rojecki
Team: Vipul Dhariwal, Zahra Fatemi, Andrew Wentzel, Lauren Levine
Funding: NSF grant #2031095
The rapid spread of the novel SARS-CoV-2 coronavirus has paralyzed societies and strained health care systems, with a rising death toll and severe economic consequences. Stay-at-home orders around the globe have been embraced by some and protested by others. At the same time, little is known about the spectrum of attitudes towards these orders and people’s justifications for following or resisting them. This research will develop algorithms for analyzing stay-at-home attitudes on social media, connecting these attitudes to pandemic impact through a novel visual representation that takes geographical location and socioeconomic context into account. This research will bring greater awareness to the public about the role of values that protect life in public discourse and their influence on citizens’ appraisal of policies that affect their own well-being. Shared values, beliefs, and understandings build the social cohesion and cooperation needed to build greater economic prosperity. The results of this project will help policy makers craft more persuasive public health directives to enhance public health.
Framing–highlighting certain aspects of an issue or event–can have a significant impact on the formation of perspective. To address the complexity of modern information networks, this project will develop algorithms that automatically detect frames propagated through social media. It focuses on value frames because people use those values to justify a position and issues can be re-framed accordingly to appeal to and change the opinions of target audiences. Creating the first dataset of its kind, this project will collect and annotate tweets that include stay-at-home attitudes and core value frames. Current state-of-the-art approaches to detecting attitudes in tweets are limited to binary or ternary classification of either sentiment (positive, negative, neutral) or language type (abusive versus “normal”). By bringing together state-of-the-art deep learning models with models that have more explanatory power, this project will devise a novel methodology for identifying values frames in microblogs that takes advantage of semantic and discourse structure information. By enabling statistical computing and analysis over geospatial data, for which few techniques exist currently, this research will make it possible to analyze datasets at multiple levels of spatial aggregation and to compare temporal and spatial differences to enhance participation and promote positive public health outcomes.