The Anatomy of Online Video Popularity

Date: 29/9/21 1PM AEST

Speaker: Prof. Lexing Xie http://users.cecs.anu.edu.au/~xlx/

Title: The Anatomy of Online Video Popularity

Recording: https://csiro.webex.com/webappng/sites/csiro/recording/c58d035f02ff103a93e7005056ba0435/playback

Bio: Lexing Xie is Professor of Computer Science at the Australian National University, she leads the ANU Computational Media lab (http://cm.cecs.anu.edu.au). Her research interests are in machine learning and the social web. Of particular recent interests are stochastic time series models, neural networks for sequences and graphs. Her research is supported by the US Air Force Office of Scientific Research, Data61, Data to Decisions CRC and the Australian Research Council. Lexing’s research has received seven best paper awards and honourable mentions in ACM and IEEE conferences between 2002 and 2019. Before the ANU, she was Research Staff Member at IBM T.J. Watson Research Centre in New York, and adjunct assistant professor at Columbia University. She received BS from Tsinghua University, Beijing, China, and PhD from Columbia University, all in Electrical Engineering.

Abstract: What makes a video popular, and what drives collective attention online? This talk gives an overview to our recent work in understanding and predicting online collective attention, especially for YouTube videos. I will first describe longitudinal measurement studies on video popularity history, the networks among videos, and quantifying popularity vs engagement. I will then discuss a physics-inspired stochastic time series model that connects exogenous stimuli and endogenous responses to explain and forecast popularity. This, in turn, leads to a set of novel metrics for forecasting expected popularity gain per share, and sensitivity to promotions. Finally, I will describe a few novel machine learning models that connects self-exciting point processes to epidemics, that derives a succinct and general representation for point processes, and one that models networks of time series. These results will guide our further pursuits on understanding and modelling the attention economy.