2025 Theses Doctoral

Creator Content Production and Promotion Decisions on Online Video Platforms

Yang, Yu Ming

This dissertation explores important substantive issues in the domains of digital markets and creator economy, through integrating new types of data, such as high-dimensional, unstructured data and novel, high-frequency data with machine learning tools, statistical/econometric methods and insights from behavioral science. In the first essay, I study what makes for a good thumbnail and how a thumbnail, relative to its video content, affects viewers’ video behavior. I propose a video mining procedure that decomposes high-dimensional video data into interpretable and consumer-relevant features (image content, aesthetics and affective emotions), leveraging computer vision, deep learning, text mining and advanced large language models.

Motivated by behavioral theories such as expectation-disconfirmation theory and Loewenstein's theory of curiosity, I construct theory-based measures to evaluate the thumbnail relative to the video content to assess the degree to which the thumbnail is representative of the video. Using both secondary data from YouTube and a novel video experimental platform called “CTube” (a simplified version of YouTube) that I create to exogenously randomize thumbnails across videos, I find that aesthetically pleasing thumbnails lead to overall positive outcomes across measures. However, content disconfirmation between the thumbnail and the video can lead to opposing effects - viewers may feel dissatisfied when there is disconfirmation between the thumbnail and the video, but once viewers observe the thumbnail, dissatisfaction disappears, leaving only the positive information-seeking effect. In the second essay, I further explain the behavioral processes underlying the thumbnails by building a Bayesian learning model in a high-dimensional video context. I postulate that thumbnails may affect video reactions through the roles of expectation-based reference points (which shape viewers’ expectations for content prior to watching a video) or informational reference points (which build anticipation for the upcoming video content).

These two roles can create opposing effects on consumers’ video viewing behavior. I model consumers' decisions to click on a video and continue watching the video as based on their priors (the thumbnail) and updated beliefs of the video content (the video's frames, characterized as multi-dimensional and correlated video topic proportions). Leveraging the high-frequency clickstream data tracked by my “CTube” platform, I estimate the Bayesian learning model and suggest that viewers overall prefer watching videos longer when there is a higher disconfirmation between their initial content beliefs based on the thumbnail and updated beliefs based on the observed video scenes (signals). In addition, viewers prefer less content disconfirmation before observing the thumbnail, highlighting that the role of disconfirmation may change before and after viewers observe the moment highlighted by the thumbnail. Based on the model's estimates, I then run a series of counterfactual analyses to propose optimal thumbnails given creators' different objectives and compare them with current practices of thumbnail recommendation to guide creators and platforms in thumbnail selection.