Melly

• 

  Slide 2

• 

  Slide 4

• 

  Slide 3

• 

  Slide 1

• 

  Slide 5

• 

  Slide 6

Team

Michael Wagman mlwagman@gmail.com MIT physics postdoc

Kelly Buckley kelly.buckley1113@gmail.com Microsoft Research research assistant

Inspiration

Data-driven music lovers trying to predict the future

What it does

Predicts music data including sales, listens, video views, and Shazams for any song in the BuzzAngle database.

How we built it

We built a mathematical model of the relationships between sales, listens, video views, and Shazams, trained the model parameters with part of the BuzzAngle data, and wrote a Python program that generates and visually displays the predictions for any song.

What we learned

As you can see in the graphs in the attached images, our model predicts outcomes that are similar to the ground truth. A simple model that assumes trends are constant from week-to-week does well, and a more sophisticated model that includes correlations between different features does better. This shows that there are real connections between features like Shazam and sales.

What's next for Melly

There are many ways to build even more sophisticated models (machine learning etc) and more more things to try and predict such as which songs are on track to become the next big hit.

Built With

• buzzangle
• python