Hi there, and welcome to this article! I’m going to explain how I built BeatBuddy, a web app that analyzes what you’re listening to on Spotify. Inspired by Spotify Wrapped, it aims to interpret your current mood and provide recommendations that you can tweak based on that analysis.

If you don’t want to read everything and just want to give it a try, you can do so here: BeatBuddy. For the rest, keep reading!

The Birth of the Project

I’m a data analyst and a music lover, and I believe that data analysis is a powerful way to understand the world we live in and who we are as individuals.

Music, in particular, can act as a mirror, reflecting your identity and emotions at a given moment. The type of music you choose often depends on your current activities and mood. For example, if you’re working out, you might choose an energetic playlist to motivate you.

On the other hand, if you are busy studying or focusing on crushing some data, you may want to listen to calm and peaceful music. I’ve even heard of people listening to white noise to focus, which can be described as the sound you hear when you open the windows of your car on the highway.

Another example of how music can reflect your mood is at a party. Imagine you are having a party with friends and you have to choose the music. If it’s a casual dinner, you might want to play some smooth jazz or mellow tunes. But if you’re aiming for the kind of party where everyone ends up dancing on the furniture or doing their best drunken karaoke performance of an ’80s hit, you’ll want to choose songs that are energetic and danceable. We’ll come back to these concepts in a moment.

In fact, all the music you listen to and the choices you make can reveal fascinating aspects of your personality and emotional state at any given moment. Nowadays, people tend to enjoy analytics about themselves, and it’s becoming a global trend! This trend is known as the “quantified self,” a movement where people use analytics to track their activities, such as fitness, sleep, and productivity, to make informed decisions (or not).

Don’t get me wrong, as a data nerd, I love all these things, but sometimes it goes too far — like with AI-connected toothbrushes. Firstly, I don’t need a toothbrush with a Wi-Fi antenna. Secondly, I don’t need a line chart showing the evolution of how well I’ve been brushing over the last six weeks.

Anyway, back to the music industry. Spotify was one of the pioneers in turning user data collection into something cool, and they called it Spotify Wrapped.

At the end of the year, Spotify compiles what you’ve listened to and creates Spotify Wrapped, which goes viral on social media. Its popularity lies in its ability to reveal aspects of your personality and preferences that you can compare to your friends.

This concept of how Spotify collects and aggregates data for these year-end summaries has always fascinated me. I remember asking myself, “How do they do that?” and that curiosity was the starting point for this project.

Well, not exactly. Let’s be honest: The idea to analyze Spotify data was written on a note titled “data project”-you know, the kind of note filled with ideas you’ll probably never start or finish. It sat there for a year.

One day, I looked at the list again, and with a new confidence in my data analysis skills (thanks to a year of growth and improvements of ChatGPT), I decided to pick an item and start the project.

At first, I just wanted to access and analyze my Spotify data for no particular purpose. I was simply curious to see what I could do with it.

Step 1: Getting Your Spotify Data

Starting a project like this, the first question you want to ask yourself is where the data source is and what data is available. Essentially, there are two ways to obtain your data:

1. In the privacy settings, you can request a copy of your historical data, but it takes 30 days to be delivered — not really convenient.
2. Using Spotify’s API, which allows you to retrieve your own data on demand and use different parameters to tweak the API call and retrieve various information.

Obviously, I went for the second option. To do so, you first need to create a developer project to get your API keys, and then you’re good to go.

API Response Example

Remember we talked about the fact that certain tracks are more likely danceable than others. As human beings, it’s quite easy to feel if a song is danceable or not — it’s all about what you feel in your body, right? But how do computers determine this?

Spotify uses its own algorithms to analyze every song in its catalog. For every song, they provide a list of features associated with it. One use of this analysis is to create playlists and give you recommendations. The good news is that their API provides access to these analyses through the audio_features endpoint, allowing you to access all the features of any song.

For example, let’s analyze the audio features of the famous song “Macarena,” which I’m sure everyone knows. I won’t cover every parameter of the track in detail, but let’s focus on one aspect to better understand how it works — the danceability score of 0.823.

According to Spotify’s documentation, danceability describes how suitable a track is for dancing based on a combination of musical elements, including tempo, rhythm stability, beat strength, and overall regularity. A score of 0.0 is the least danceable, and 1.0 is the most danceable. With a score of 0.823 (or 82.3%), it’s easy to say that this track is very danceable.

The Three Temporalities

Before going further, I need to introduce a concept with the Spotify API called time_range. This interesting parameter allows you to retrieve data from different time periods by specifying the time_range:

• short_term: the last 4 weeks of listening activity
• medium_term: the last 6 months of listening activity
• long_term: the entire lifetime of your listening activity

Let’s illustrate this with an example: if you want to get your top 10 tracks from the last 4 weeks, you can call the corresponding endpoint and pass the time_range as a parameter like this : https://api.spotify.com/v1/me/top/artists?time_range=short_term&limit=10

Calling this will give you your top 10 artists from the past month.

Step 2: Interpreting the Results

With all this information available, my idea was to create a data product that allows users to understand what they are listening to, and to detect variations in their mood by comparing different temporalities. This analysis can then show how changes in our lives are reflected in our music choices.

For example, I recently started running again, and this change in my routine has affected my music preferences. I now listen to music that is faster and more energetic than what I typically listened to in the past. That’s my interpretation, of course, but it’s interesting to see how a change in my physical activity can affect what I listen to.

This is just one example, as everyone’s musical journey is unique and can be interpreted differently based on personal experiences and life changes. By analyzing these patterns, I think it is pretty cool to be able to make connections between our lifestyle choices and the music that we like to listen to.

Making Data Insight Accessible

The deeper I got into this project, the more I came to realize that, yes, I could analyze my data and come to certain conclusions myself, but I wanted everyone to do it.

To me, the simplest way to share data insights with non-technical people and make it so very accessible is not through a fancy BI dashboard. My idea was to create something universally accessible, which led me to develop a mobile-friendly web application that anyone could use.

To use the app, all you need is a Spotify account, connect it to BeatBuddy with the click of one button, and you’re done !

Measuring Musical Emotions

Let’s look at another feature of the app: measuring the happiness level of the music you’re listening to, which could reflect your current mood. The app aggregates data from your recent top tracks, focusing on the ‘valence’ parameter, which represents musical happiness, with 1 being super happy music. For instance, if the average valence of your current tracks is 0.432, and your all-time average is 0.645, it might suggest a shift towards more melancholic music recently.

However, these analyses should be taken with a grain of salt, as these numbers represent trends rather than absolute truths. Sometimes, we shouldn’t always try to find a reason behind these numbers.

For example, if you were tracking your walking pace and discovered you have been walking faster lately, it doesn’t necessarily mean you’re in more of a hurry — it could be due to various minor factors like changes in weather, new shoes, or simply a subconscious shift. Sometimes changes occur without explicit reasons, and while it is possible to measure these variations, they do not always require straightforward explanations.

That being said, noticing significant changes in your music listening habits can be interesting. It can help you think about how your emotional state or life situation might be affecting your musical preferences. This aspect of BeatBuddy offers an interesting perspective, although it’s worth noting that these interpretations are only one piece of the complex puzzle of our emotions and experiences

Step 3: Take Data-Driven Decisions

Let’s be honest, analyzing your listening habits is one thing, but how do you take action based on this analysis? In the end, making data-driven decisions is the ultimate goal of data analysis. This is where recommendations come into play.

Recommendations Based on Your Selected Mood

An interesting feature of BeatBuddy is its ability to provide music recommendations based on a mood you select and the music you like.

For instance, you might realize that what you are listening to has a score of 75% popularity (which is quite high), and you want to find hidden gems tailored to your tastes. You can then tweak the “Popularity” slider to, say, 25% to create a fresh playlist with an average score of 25% popularity.

Behind the scenes, there’s an API call to Spotify’s algorithm to create a recommendation based on the criteria you’ve selected. This call generates a playlist recommendation tailored to both your personal tastes and the mood parameters you’ve set. It uses your top 5 recent tracks to fine-tune Spotify’s recommendation algorithm according to your choices.

Once you’re happy with the playlist, you can save it directly to your Spotify library. Each playlist comes with a description that details the parameters you chose, helping you remember the mood each playlist is meant to evoke.

Final Thoughts

Developing a web application that analyzes Spotify data has been a challenging but rewarding journey. I have been pushed out of my comfort zone and gained knowledge in several areas, including web API, cookie management, web security, OAuth2, front-end development, mobile optimization, and SEO. Below is a diagram of the high-level architecture of the application:

My initial goal was to start a modest data project to analyze my listening habits. However, it turned into a three-month project rich in learning and discovery.

Throughout the process, I realized how closely related data analysis and web development are, especially when it comes to delivering a solution that is not only functional but also user-friendly and easily accessible. In the end, software development is essentially about moving data from one place to another.

One last note: I wanted to create an application that was clean and provided a seamless user experience. That is why BeatBuddy is completely ad-free, no data is sold or shared with any third parties. I’ve created this with the sole purpose of giving users a way to better understand their music choices and discover new tracks.

You can give the app a try here: https://www.beatbuddy.cloud

If you have any comments or suggestions, I’m all ears! Your feedback is really important.

For those interested in a deeper dive, keep an eye out for my upcoming article.

Cheers!

Lazare