Unbox the Blackbox: Predict and Interpret YouTube Viewer Using Deep Learning

Reporter: XieJiaheng Assistant Professor University of Delaware

Time: 9:00-10:30 am, May 16, 2022

Tencent conference number: 929 814 766

Introduction to the report:

As video-sharing shapes an emerging social media landscape, content creators and businesses urge to prioritize video viewership prediction to optimize influence and marketing outreach with minimum budgets. Although deep learning champions viewership prediction, it lacks interpretability, which is required by regulators and is fundamental to guiding video production and accepting predictive models. Existing interpretable predictive models face the challenges of imprecise interpretation and negligence of unstructured data. Following the design-science paradigm, we propose a novel information system, Precise Wide-and-Deep Learning (PrecWD), that accurately predicts viewership leveraging unstructured raw videos and well-established features while precisely interpreting feature effects. PrecWD outperforms benchmarks in two contexts – health video and misinformation viewership prediction – and achieves superior interpretability in a user study. We contribute to IS knowledge base by enabling precise interpretability in video-based predictive analytics. We also contribute to IS design theory with generalizable design principles in model development. Our system and findings are deployable to improve video-based social media presence.

Brief introduction of the reporter:

Dr. XieJiaheng is an assistant professor in the Department of Accounting and Management Information Systems, Alfred Lerner School of Business, University of Delaware. He received his doctorate from the University of Arizona's Eller School of Management. His research interests include in-depth learning, health risk analysis and business analysis. His previous work has been published in many important journals, including MIS Quarterly, Journal of Management Information Systems, and Journal of American Medical Information Association.

(Undertaken by: Department of Management Engineering, Scientific Research and Academic Exchange Center)