The rising demand for online video content has fostered one of the fastest-growing markets as evidenced by the popularity of platforms like TikTok. Because video content is often difficult to create, platforms have attempted to leverage recent advancements of artificial intelligence (AI) to help creators with their video creation process. However, surprisingly little is known about the effects of AI on content creators’ productivity and creative patterns in this emerging market. Our paper investigates the adoption impact of AI-generated voice – a generative AI technology creating acoustic artifacts – on video creators by empirically analyzing a unique dataset of 4,021 creators and their 428,918 videos on TikTok. Utilizing multiple audio and video analytics algorithms, we detect the adoption of AI voice from the massive video data and generate rich measurements for each video to quantify its characteristics. We then estimate the effects of AI voice using a difference-in-differences model coupled with look-ahead propensity score matching. Our results suggest that the adoption of AI voice increases creators’ video production and that it induces creators to produce shorter videos with more negative words. Interestingly, creators produce more novel videos with less self-disclosure when using AI voice. We also find that AI-voice videos received less viewer engagement unintendedly. Our paper provides the first empirical evidence of how generative AI reshapes video content creation on online platforms, which provides important implications for creators, platforms, and policymakers in the digital economy.
With the increasing use of online matching platforms, predicting matching probability between users is crucial for efficient market design. Although previous studies have constructed various visual features to predict matching probability, facial features, which are important in online matching, have not been widely used. We find that deep learning-enabled facial features can significantly enhance the prediction accuracy of a user’s partner preferences from the individual rating prediction analysis in an online dating market. We also build prediction models for each gender and use prior theories to explain different contributing factors of the models. Furthermore, we propose a novel method to visually interpret facial features using the generative adversarial network (GAN). Our work contributes to the literature by providing a framework to develop and interpret facial features to investigate underlying mechanisms in online matching markets. Moreover, matching platforms can predict matching probability more accurately for better market design and recommender systems.
Park, Sungho, Gene Moo Lee, Donghyuk Shin, Sang-Pil Han. “When Does Congruence Matter for Pre-roll Video Ads? The Effect of Multimodal, Ad-Content Congruence on the Ad Completion“, Working Paper [Last update: Jan 29, 2023]
Previous title: Targeting Pre-Roll Ads using Video Analytics
Funded by Sauder Exploratory Research Grant 2020
Presented at Southern Methodist University (2020), University of Washington (2020), INFORMS (2020), AIMLBA (2020), WITS (2020), HKUST (2021), Maryland (2021), American University (2021), National University of Singapore (2021), Arizona (2022), George Mason (2022), KAIST (2022), Hanyang (2022), Kyung Hee (2022), McGill (2022)
Research assistants: Raymond Situ, Miguel Valarao
Pre-roll video ads are gaining industry traction because the audience may be willing to watch an ad for a few seconds, if not the entire ad, before the desired content video is shown. Conversely, a popular skippable type of pre-roll video ads, which enables viewers to skip an ad in a few seconds, creates opportunity costs for advertisers and online video platforms when the ad is skipped. Against this backdrop, we employ a video analytics framework to extract multimodal features from ad and content videos, including auditory signals and thematic visual information, and probe into the effect of ad-content congruence at each modality using a random matching experiment conducted by a major video advertising platform. The present study challenges the widely held view that ads that match content are more likely to be viewed than those that do not, and investigates the conditions under which congruence may or may not work. Our results indicate that non-thematic auditory signal congruence between the ad and content is essential in explaining viewers’ ad completion, while thematic visual congruence is only effective if the viewer has sufficient attentional and cognitive capacity to recognize such congruence. The findings suggest that thematic videos demand more cognitive processing power than auditory signals for viewers to perceive ad-content congruence, leading to decreased ad viewing. Overall, these findings have significant theoretical and practical implications for understanding whether and when viewers construct congruence in the context of pre-roll video ads and how advertisers might target their pre-roll video ads successfully.
We study the spillover effects of the online reviews of other covisited products on the purchases of a focal product using clickstream data from a large retailer. The proposed spillover effects are moderated by (a) whether the related (covisited) products are complementary or substitutive, (b) the choice of media channel (mobile or personal computer (PC)) used, (c) whether the related products are from the same or a different brand, (d) consumer experience, and (e) the variance of the review ratings. To identify complementary and substitutive products, we develop supervised machine-learning models based on product characteristics, such as product category and brand, and novel text-based similarity measures. We train and validate the machine-learning models using product pair labels from Amazon Mechanical Turk. Our results show that the mean rating of substitutive (complementary) products has a negative (positive) effect on purchasing of the focal product. Interestingly, the magnitude of the spillover effects of the mean ratings of covisited (substitutive and complementary) products is significantly larger than the effects on the focal product, especially for complementary products. The spillover effect of ratings is stronger for consumers who use mobile devices versus PCs. We find the negative effect of the mean ratings of substitutive products across different brands on purchasing of a focal product to be significantly higher than within the same brand. Lastly, the effect of the mean ratings is stronger for less experienced consumers and for ratings with lower variance. We discuss implications on leveraging the spillover effect of the online product reviews of related products to encourage online purchases.
Presented in WITS (Auckland, New Zealand 2014), and WISE (Auckland, New Zeland 2014), HICSS (Kauai, HI 2016)
Dissertation Paper #2
This article studies the strategic network formation in a location-based social network. We build an empirical model of social link creation that incorporates individual characteristics and pairwise user similarities. Specifically, we define four user proximity measures from biography, geography, mobility, and short messages. To construct proximity from unstructured text information, we build topic models using Latent Dirichlet Allocation. Using Gowalla data with 385,306 users, 3 million locations, and 35 million check-in records, we empirically estimate the model to find evidence on the homophily effect on network formation. To cope with possible endogeneity issues, we use exogenous weather shocks as our instrumental variables and find the empirical results are robust: network formation decisions are significantly affected by our proximity measures.
This paper studies strategic network formation in a location-based social network. We build a structural model of social link creation that incorporates individual characteristics and pairwise user similarities. Specifically, we define four user proximity measures from biography, geography, mobility, and short messages. To construct proximity from unstructured text information, we build topic models using latent Dirichlet allocation. Using Gowalla data with 385,306 users, three million locations, and 35 million check-in records, we empirically estimate the structural model to find evidence on the homophily effect in network formation.