
Massive Language Versions as Facts Interfaces for Overall health Purposes
Abstract
Social media has grow to be a crucial channel for persons to request and share facts, even on delicate subject areas these kinds of as healthcare assistance and own ordeals connected to their wellbeing and wellbeing. Even so, the mainly unvetted nature of social media may well inadvertently expose users to medical mis- and disinformation. On the other hand, the huge amounts of health and fitness-connected discussions accessible online provides opportunities for epidemiologists and general public well being specialists, e.g. to observe the prevalence of sicknesses and tell public well being interventions. Having said that, present social media examination strategies often ignore vital contextualizing facts these as visuals shared together with created posts and personal attributes of the people (e.g., demographics). In flip, this can lead to suboptimal and biased inferences, especially for underrepresented groups.
In this chat, Silvio Amir will current my ongoing do the job towards addressing these difficulties. Initial, he will discuss novel strategies and datasets to determine health-related statements on social media and retrieve peer-reviewed literature to help/refute said promises. He will then display how to lessen guide initiatives in deploying these programs in other domains by eliciting annotations from significant language models. Second, He will focus on the chances and difficulties of incorporating illustrations or photos and private attributes in products for longitudinal social media information. Then, he will present a circumstance-examine on the application of personalized multimodal designs to estimate the prevalence of many mental health circumstances this kind of as depression, stress, and PTSD from social media.
Biography
Silvio Amir is an assistant professor in the Khoury University of Laptop or computer Sciences. His investigate develops All-natural Language Processing and Equipment Discovering techniques for individual and consumer generated text, these types of as social media and scientific notes from Electronic Well being Documents. Amir is primarily intrigued in approaches for jobs involving subjective, customized or person-level inferences (e.g., viewpoint mining and digital phenotyping). In certain, his function aims to strengthen the trustworthiness, interpretability and fairness of predictive models and analytics derived from personal and person created facts. His analysis is aspect of ongoing endeavours to create Human-centered AI (i.e., to empower alternatively than exchange human beings) and AI for Social Fantastic (i.e., to tackle meaningful social, societal, and humanitarian challenges). To obtain these ambitions, he normally collaborates with domain industry experts in multidisciplinary assignments to deal with serious-environment challenges in the social sciences, medicine, and epidemiology.
Amir gained his doctorate from the College of Lisbon, conducting portion of his doctoral investigate as a browsing researcher at the University of Texas at Austin and at Northeastern University in Boston. He then moved to John Hopkins College, where he accomplished his postdoctoral exploration in the Center for Language and Speech Processing and served as a lecturer at the Whiting College of Engineering.