Risk and strength: determining the impact of area-level racial bias and protective factors on birth outcomes

Univ Of Maryland, College Park, College Park MD

Investigators

Linked publications & trials

Abstract

PROJECT SUMMARY Our parent grant focused on using deep machine learning models to analyze the text of social media data to create area-level measures of racial sentiment and to link them with birth outcomes. We focused solely on race-based biases and did not analyze visual representations of bias. The promising findings from this work open up questions about the generalizability of this approach for investigating additional health disparities across other intersecting identities like gender and sexual orientation. We also seek to develop new AI/ML methods to derive sentiment from image-based content, which is an increasingly common form of communication on social media. The administrative supplement will allow us to advance this AI/ML area by making the data AI/ML ready for image analysis, training multimodal models that incorporate both text and image analysis, and creating valuable resources that will shorten the time and specialized expertise needed to implement machine learning models to investigate the impact of area-level biases on health inequities. We will be sharing our AI/ML-ready social media-derived measures by creating a public geoportal for interactive data visualization and sharing, which will contain repositories of aggregated data that will facilitate the use of this big data source for future applications in health equity research.