Face Value: Ph.D. Student Flora Feng Uses AI to Measure Charisma and Uniqueness

Doctoral student Flora Feng studies how artificial intelligence (AI) can quantify and predict intangible qualities like charisma and visual uniqueness.

Xiaohang (Flora) Feng is in her final year as a doctoral candidate in marketing at the Tepper School of Business. After graduation, she plans on staying in academia to pursue her research and scholarship. Her research applies technological advancements such as computer vision (AI that allows computers to derive information from visual media such as images or video), generative AI, and machine learning to marketing problems, focusing on explainable AI and sustainability. She combines game theory, causal inference, and structural models to analyze mechanisms in consumer behavior. 

Flora is the lead author of the paper, “An AI Method to Score Celebrity Visual Potential from Human Faces,” published in the Journal of Marketing Research. She and her co-authors examine the core concept of charisma and how it relates to celebrity. Charisma is an important trait that inspires devotion, trust, and attention from others. Starting with the hypothesis that facial features can predict the visual aspects of charisma, the researchers employed a machine-learning model to analyze 6,000 images of celebrities and 6,000 non-celebrities, identifying 11 facial features to develop a metric called celebrity visual potential, or CVP. 

With an accuracy rate of almost 96 percent, the model found that large eyes, darker skin tones, and strong sexual dimorphism correlated with higher celebrity potential, while babyfaced features, narrow jaws, and overall facial averageness correlated with lower CVP scores. The CVP can be used in a variety of ways. For example, it may be a useful tool for screening photos such as those to be used in advertisements. There also may be applications to spokespeople or talent agencies in terms of makeup and hairstyles and selecting the right social media influencers to promote products. 

While the first study applies machine learning to predict human desirability, the second, “Visual Uniqueness in Peer-to-Peer Marketplaces: Machine Learning Model Development, Validation, and Application” (conditionally accepted by the Journal of Consumer Research at the time of this writing) applies it to objects and spaces. Specifically, it examines how the visual uniqueness of an Airbnb listing impacts its success. They used a machine-learning model to analyze 481,747 images of Airbnb listings in New York City. The model’s predictions of uniqueness aligned with how humans perceived uniqueness. When looking at how a listing’s uniqueness affected demand, Flora and her co-authors found that demand increased with uniqueness up to a point, and after that point, demand decreased as uniqueness increased, showing the relationship between uniqueness and demand as an inverted U.

The uniqueness model can help companies know how to market their products. While the paper focused on Airbnb, the model can be applied to other situations, such as e-commerce. For example, a retailer can highlight listings that had the highest uniqueness, and also use this to help sellers know how to change their listings to increase demand. Another application is in the design of search and recommendation algorithms. If a company like Airbnb knows a customer is looking for a unique experience, it can prioritize listings high in uniqueness. 

Properties that were too generic or too unconventional performed worse than those with a moderate level of uniqueness. Hosts with strong ratings or quick response times saw less negative impact from extreme uniqueness. These findings offer practical guidance for Airbnb hosts. Properties should balance distinctiveness with familiarity. A space that is too ordinary may fail to attract attention, while one that is too unusual may make potential guests hesitant. The study also highlights how uniqueness interacts with trust. A host with strong reviews can take more risks with design, while a new host may benefit from a more conventional look.

Both studies, at their core, are about the quantification of intangibles. Charisma and uniqueness have historically been subjective, amorphous qualities. Now, they are data points that can be measured and optimized. 

In addition to these two papers, Flora is revising a paper for Marketing Science and has authored a chapter on image analytics in marketing in Artificial Intelligence in Marketing and another chapter on AI pricing in The Elgar Encyclopedia of Pricing.