Disney’s Deep Dive on Personality Research, and Its Potential Implications for Brand Marketers

Share this:

A few of Mickey Mouse’s pals within a division of The Walt Disney Company have been working on a method to better target marketing by learning more about consumers’ personalities.

The research labs division at the entertainment giant was established to develop ideas that the company might consider exploring, and  well as share some of its findings with the public. Some of the focus at Disney Research is on high profile areas such as robotics and computer graphics. Meanwhile, research scientist Maarten Bos has worked on ways to better understand audience personality and how ads can be tailored to them based on the images used. Bos spoke to Street Fight about how his team of behavioral scientists conducted a series of studies on how people react to images, and ways that information might be used within Disney and beyond.

What was the basis of the study and what did it reveal about how people might react to ads?
These are all concepts to figure out new approaches. What we are doing is look at whether we can predict what kind of person likes what kind of image. I’m talking about personality type. An introverted person or an extroverted person, what kind of image might they like? When I say what kind of image, we analyze images with over 200 features like contrast or the content of the image itself. We use some automated methods to extract all those features. Then we try to map those things on to each other. We try to map each personality onto image features.

If we know your personality, we can then tailor a marketing message to you with the image that we want to send to you. We had a lot of users online rate their preferences for certain images — from more than 1,000 images. We get their preferences, measure their personality, and then extract the features automatically and make a model of all of that. With that we try to find out: “What if we showed new pictures to a new group of people, that we know the personality of, can we predict who will like which image?” And we were successful in that.

These effects are small; these are tiny effects but if you market at scale, then those small effects can have an impact. At the very least you can try to not send an image that someone might not like.

The big “if” is how do you know people’s personality? There’s a way to do that by looking at people’s [publicly shared] Facebook “likes.” There’s a guy at Stanford who worked on this model when he was at the University of Cambridge, Michal Kosinski — he built a model of personality based on Facebook “likes.” From a certain number of Facebook “likes” from people you can actually predict their personality. That is one way to do it.

Personalization is hard because it is expensive. If you want to do this like a boutique hotel does, they charge you more because it’s hard to make a service personal. That doesn’t have to be the case if you can automate it. To do this at scale, you have to automate it through methods like feature extraction from images. Putting those things together with automated personality typing, that can be an approach to go with.

As far as I know, Disney doesn’t use personality types in marketing yet. As far as I know, they don’t use Facebook “likes” in any way. But more importantly, this is not about personality itself. This is one vehicle to do segmented marketing. What’s interesting is the idea of the model. We have some features of people; we have some features of pictures — we can machine learn those two things together.

The images used to get these features and preferences — were these stock images aimed at trying to represent different personalities? Were they keyed to any brand items or products?
These are from a database; anyone can find them. In the end, you can apply this method to any images you select. You just have to copy our methods. It’s not a super-new approach, necessarily. People have done things like that before. One of my co-authors on the paper, she used a database that contains images that people favorited on Flickr. That’s self-reported and attributed personality scores for Flickr users. What she did was stitch those things together. We also used images from Shutterstock, which has predefined categories like nature or buildings or people.

What are the potential applications for this? Where could this method of personalization come into play for targeted and local marketing in the real world?
It’s the hardest question to answer because it is very speculative. In any situation where you can extract features or find certain features you have about people — if you localize by city or any features you know about a group of people, you can start building a model just like this. Personality is just a vehicle, but it’s not the only way to segment people. If you have two cities, and one of them has a more heavily Spanish-speaking population and the other has a more heavily English-speaking population, or there are different combinations of the two, that can be interesting. Those are specific features about people.

If you are targeting for an athletic club that is for women or for men, you could tailor your advertising for that. People have been doing that forever. It can become more predictive if you use this approach of having extracted features, having a list of things you know about a person, and then combine them. In a very broad sense, it could work for any segment. You just have to test it out.

Disney Research’s Maarten Bos

Does this open up the possibility to use other types of features and metrics to understand your audience’s personality?
We used a very established personality model, The Big Five. That is one way to do this. There are sorts of interesting personality features out there. There’s people’s need for closure and touch, all different ways to slice up different segments.

Are we at a point now, with your research and others, where marketing can explore radical new options in how audience is understood and approached?
Definitely. If you think about it, Google has been trying to tailor advertising forever because they want to show you the right images so you don’t get annoyed with them. There’s been A/B testing for a very long time. The reason it can be effective is because at some point you can automate it. Personalizing is expensive; you don’t want to have to sit down for every single person and ask which image you can send to them that will attract them to your advertising. That’s not feasible. What opens the door here is computer science, automated methods of extracting features. It’s not even stealing a job from someone; it was impossible to do before.

Joao-Pierre Ruth is a Street Fight contributor.