Premise

Those of have ever tried planning songs for someone else, they know the pressure we;re talking about. It is a tricky path to navigate. If you know the person well, it's a bit easier but let's be honest, playing for a group of people is pretty much a nightmare. Which is when we fallback on trysted Spotify playlists.

But that is not really the best solution, is it.

<aside> ❓ How might we get a group of people to vibe together as they listen to music based on their individual interests.

</aside>

Brief

• The project is done for the course - Designing Smart Systems.
• The learning outcomes for the course were:
  • Identify opportunities for applications of AI and Data science techniques.
  • Basic understanding of cleaning and sorting through a dataset and perform EDA on it.
  • Design of systems that learn and adapt as it gathers more information over time.

Research

Existing Solutions

• Spotify Blend: Blend proposes to match and mix (only) two playlists. It provides a percentage match of music taste and also gives playlist that is unique to your blend. In the image below, we have screenshots of the “Liked songs” playlists of two members from our team.

  

What Spotify Blend does is just sandwich these playlists. Clearly, these songs fall on far apart spectrum of what either team member of would listen to.

• Musictaste.net: This third party platform gives a music taste match but fails to give a combined playlist. Additionally, it needs to keep syncing with Spotify to update the users listening preferences and potentially shows some privacy issues.

Process

Existing Solutions

GrooveSync uses the liked song playlist of the users in a group and evaluates the compatibility of other songs and playlists to the group and recommends songs that everyone can enjoy together

Execution

• As a team we combined all our liked songs playlists and matched it with the music factors that Spotify uses in their current algorithm.
• EDA: To understand the data better, identify outliers and see correlations.
• We used methods like Gaussian Mixture Models clustering, custom clustering and NeighbourAssess to generate that smart playlist.
• We then used the algorithm on the Top 100 playslist to see if it can sleect songs that all the members on the team could vibe to.

Reports and Presentation

Presentation

The presentation outlines the problems and the methods and processes we used to solve it.

GrooveSync_Presentation.pdf

Report

A fairly detailed version going into details of the clustering and NeighbourAssess methods.

GrooveSync_Report.pdf