The Discontinuous Diffusion of User-Generated Content

Abstract

This research investigates the discontinuous diffusion process associated with User Generated Content (UGC). Specifically, we examine the daily views for a sample of videos posted on YouTube. We assume that the audience for each video is drawn from two segments that represent the initial poster’s segment and a secondary diffusion segment. The UGC spreads within each segment according to segment-specific Weibull diffusion processes and across segments according to a stochastic process that allows the second segment to begin its diffusion process at a later point in time.

Our results indicate that UGC’s diffusion in many cases is discontinuous, and the secondary diffusion segment, on average, begins its adoption process significantly later than the initial segment. We also examine two factors that may impact diffusion and diffusion discontinuity: seeding and content. We find that while seeding through posters who have large observe network sizes (such as subscribers or friends) may generate more immediate viewership, seeding through posters who have large, unobserved network sizes (individuals who are connected to the poster but unobservable), relative to the secondary segment, reduces diffusion discontinuity. In addition, we find that the magnitude of discontinuity varies across video content. Furthermore, we provide managers with a methodology to forecast the timing of discontinuity.

Keywords: social media, user-generated content, probability model, mixture model, Weibull distribution, Hierarchical Bayes, diffusion discontinuity, information chasm

JEL Classification: M30, M31