Creating Brand Personas with Machine Learning by@jzeoli
5,393 reads

Creating Brand Personas with Machine Learning

Read on Terminal Reader

Too Long; Didn't Read

Companies Mentioned

Mention Thumbnail
Mention Thumbnail
featured image - Creating Brand Personas with Machine Learning
Joe Zeoli HackerNoon profile picture


Joe Zeoli
react to story with heart

As a developer, I don’t usually sit in on our brand strategy meetings, but seeing as this particular meeting was about 20nine’s brand, I was able to get a taste of how they work. One of our assignments was to come up with a character (real or fictitious) that best represented the personality of the company.

Around the same time, I attended a local conference and got to sit in on a talk that was about predicting a user’s consumption habits & preferences based on their Twitter activity. Afterwards, I looked into the science behind the API that made this possible — IBM Watson’s Personality Insights. The Personality Insights service infers personality characteristics from content inputs based on an open-vocabulary approach. It was trained on thousands of personality test scores along with their corresponding Twitter accounts.

My hypothesis was that by scraping a company’s webpage we could get a pretty good idea of the brand’s “personality” and match that with a celebrity to create a persona. I was actually able to find this sample application that would help to create this:

About The Profiles

Beyond knowing the API, I had to figure out exactly what information was inferred based on the data that was given. Watson’s Personality Insights uses three well-known models for determining personality characteristics.

Big Five

The Big Five is the most widely used model for generally describing personality characteristics and how a person engages with the world. This model has 5 dimensions: Agreeableness, Conscientiousness, Extraversion, Emotional range and Openness.


These describe which aspects of a product will resonate with a person. The model includes twelve characteristic needs: Excitement, Harmony, Curiosity, Ideal, Closeness, Self-expression, Liberty, Love, Practicality, Stability, Challenge, and Structure.


These describe motivating factors that influence a person’s decision making. The model includes five values: Self-transcendence / Helping others, Conservation / Tradition, Hedonism / Taking pleasure in life, Self-enhancement / Achieving success, and Openness to change / Excitement.

You can read more about these here:

The Build

I manually grabbed about 150 celebrity Twitter accounts to analyze. Using the Twitter API, I grabbed the celebrity’s tweets and sent it through the Personality Insights API. The output was saved to a MongoDB instance to later compare with a brand.

Next, I built a very simple website scraper in node.js that grabbed the top-level URLs and scrapped all heading, paragraph and span tags from the pages. Since it only grabs the top-level URLs, inputting the homepage URL and the Blog URL might return different results since the extra blog pages would be picked up on the latter. The content gets grouped together and analyzed by Watson.

We can then compare any website to the celebrities we saved earlier.


Example output for

To take it a step further I also grabbed around 150 of the world’s largest brands’ Twitter accounts and used the same mechanism to compare against.


Example output for

Comparing All Brands

An unexpected consequence of creating this tool is the ability to compare and contrast the saved brands with each other. They can be filtered by industry and analyzed to potentially spot trends and find gaps in personas for a specific market.

Below are examples of where these brands fall in regards to challenge, imagination and practicality.




It’s really fascinating to see companies like Disney and MTV score high for imagination while Red Bull scores high for challenge. If, for example, we wanted a brand to be more on the imaginative side, we could research the content Disney puts out to skew more towards that personality characteristic.

Machine learning is being utilized for more and more challenges so it was fun to take on part of the branding process with math and data.

Interested to see the tool I built? You can check it out here:


. . . comments & more!
Hackernoon hq - po box 2206, edwards, colorado 81632, usa