Joseph Johnson, Gang Ren, Mitsunori Ogihara, and Hyunhwan “Aiden” Lee, “DEEPSENSE: A Deep-Learning Predictive Tool for Evaluating Effectiveness of Video Commercials,”
accepted for Theory + Practice in Marketing Conference 2019
Revising manuscript for Journal of Marketing.
Artificial Intelligence in Management Conference, University of Southern California. March 14, 2020
Testing and validation of television commercials require sophisticated audience response collection schemes at various production and deployment stages of a commercial’s lifecycle including script writing, actor casting and footage shooting. Audience response at these stages allows interactive adjustment to the production and deployment processes which increases the effectiveness of the television commercials. Current testing advertising approaches include manual annotation, focus group studies and audience/customer opinion surveys. Besides their high operational cost, these approaches for collecting audience response data are usually time-consuming and lack immediacy.
We propose a computational “media cognition” tool that uses deep neural networks on video, audio and natural language content of commercials as training data and the audience response attributes as the training labels to form a predictive analysis framework for video commercials. Our data source is YouTube video commercials. The audience response data include the manual annotations as well as YouTube likes/dislikes and comments.
We find that the direct content modeling system performs better when we combine the video, audio, and text dimensions of the content than employing any of these dimensions alone. For predictive analysis based on a single dimension, the video stream has the highest predictive power. The central implication of our study is that firms can train deep learning algorithms directly on the content of existing commercials and use them to reliably predict audience response of unreleased media content.
Keywords: deep learning, advertising effectiveness, multimedia content analysis, long short-term memory, convolutional neural network
Hyunhwan “Aiden” Lee, Joseph Johnson, and Gerard J. Tellis. “Geo-Influence: Modeling Location-Specific Effects of Social Influence on Brand Preferences”
Revising manuscript for Journal of Marketing Research
The 41st ISMS Marketing Science Conference, University of Rome. June 20 – 22, 2019
The 2019 Haring Symposium, Indiana University. April 19 – 20, 2019
The 1st Interdisciplinary Research Cluster Day, University of Miami. April 1, 2019
The 3rd Annual Three Minute Thesis Competition, University of Miami. February 6, 2019
Behavioral Insights from Text Conference, University of Pennsylvania. January 17, 2020
Artificial Intelligence in Management Conference, University of Southern California. March 14, 2020
No brand is the market leader uniformly across geographic locations. Brands have strong and weak markets. Therefore, analysts need to track brand preferences by location. However, current methods of tracking geographic brand preferences (e.g., surveys, company sales, scanner panel data, Google Analytics, etc.) have limitations: limited sampling of brands, stores, and categories with time and region aggregation. We address these limitations through a spatial, nonparametric approach to track brand preferences at the micro-temporal and micro-geographic level using Twitter. Our approach inherits the intrinsic strengths of social media: micro-temporal, micro-geographic, faster than news media, higher frequency than sales, low cost, competitive monitoring, sentiment inclusion, and social network inclusion. A key feature of our approach is modeling the temporal geographic influence of social connections based on the followers of each post. We estimate our model for carbonated soft drinks brands. Our validation uses Pepsi sales, an apparel fashion brand, and the 2016 U.S. presidential election (a domain where marketing is critical). We find that geographic influence adds predictive power to micro-temporal and micro-geographic brand measures, leads online sales and predicts elections. Our study implies that the micro-temporal and micro-geographic social media measures provide actionable insights for marketing brands and political candidates.
Joseph Johnson, Debanjan Mitra, Sivaramakrishnan Siddarth, Hyunhwan “Aiden” Lee
Revision invited, Journal of Marketing
Existing customer-based brand equity metrics are built on customers’ choices and preferences and not on the dynamics inherent in their brand switching and timing, which potentially handicaps the prognostic ability of such metrics. To remedy this we develop a complementary metric that focuses on the rate of change in customers’ purchase transitions. We use the counting process to model the repeat and switching rates of a brand, parse out baselines for each transition rate, cumulate these baselines over a given duration, and combine them to obtain brand hazard (BH). Using customer panel data, we estimate BH for six leading brands in three product categories over sixteen two-year moving windows. We find increases in BH foretells future declines in the intercept of logit models of customers’ brand choice (BCI) – a customer based brand equity metric estimated from the same data. More important, we find that changes in BH and BCI are able to predict future changes in a brand’s price premium and revenue premium in the broader market. Our findings imply that BH complements BCI, and tracking them together provides a comprehensive prognosis of a brand’s health.
