Erik Driessen

More Than A Job

More Than A Job / 01

There are over 500 employees at Greenhouse. Even though they are challenged sufficiently with client projects, our employees are enterpreneurial, creative and inquisitive people after working hours, too. In More Than A Job we offer a stage to those Greenhousers who don’t let office walls and hours limit their passion for data, media, tech and creation. 

In the first episode we speak with Erik Driessen, Head of Digital Analytics. For his personal project Sentimentshirt he turned albums into unique designs for T-shirts through Natural Language Processing.












Can you briefly explain the concept behind Sentimentshirt?

“Sentimentshirt visualizes musical sentiment through Machine Learning. It starts with the lyrics of your favorite album. I turn every track into a datapoint through ML. Every datapoint tells something about the emotion (negative, neutral or positive) and the intensity of the emotion. In that way, every song on the album gets assigned a sentiment score. Then the scores are plotted relative to each other. By connecting those datapoints, a shape emerges that is similar to a signature. This gives you as a fan of whichever band or artist a unique T-shirt, instead of the band merch that everyone is wearing at a concert.”

Sentimentshirt is a very personal dream project. How did it start?

“Well, actually it started as a nightmare rather than a dream. I was attending a work event when a colleague showed me a heartbreaking notification. Avicii had passed away. He had committed suicide. Before that happened I had just watched the True Stories documentary, in which he announced that he was going to do more of what was best for himself. I didn’t expect him to end his life after watching that.”

Erik Driessen

More than a job / 01

How did this tragic event become the start of your project?

“I like trying out things with new technology. In this case I wanted to do something with Natural Language Processing. The Google Cloud Natural Language API allows for sentiment analysis. My hypothesis was that you could deduct Avicii’s negative mood or even depression from the lyrics on his albums. I knew that he was more involved with lyrics than other DJs.”

Was your hypothesis correct?

“No. I analyzed three albums. The first album wasn’t particularly positive or negative. The second one was rather positive. And the most recent album was very balanced.”

But then your project didn’t end.

“The fact that I couldn’t determine whether or not Avicii was depressed through analyzing lyrics with Machine Learning, only meant that the lyrics weren’t predictors. And that ML isn’t a replacement for psychologists and psychiatrists yet. But the fact that it was possible to deduce a sentiment from the lyrics was really cool to me and gave me the energy to continue doing something with it.”

What is your next step?

“The more I tell the story of Sentimentshirt, the more ideas I get for a continuation. I presented the project to some entrepreneurs and received a lot of tips to improve my pitch and website. My head also never stops working. By analyzing scripts and texts with ML it’d also be possible to convert movies and books to a Sentimentshirt.”

When you came up with the idea, did you immediately think of legally protecting it?

“No, not at all. I think it would be great if other people go and play around with it. The code is available on Github. Generally speaking, I think sharing knowledge is so much more valuable than making as much money as possible from it. However, I’m not complaining that there are more and more orders coming in at!”

Order your Sentimentshirt