Python & SpotiPY track recommendation experiment

radio_vamorda

New member
Hi,

Recently I found out a library - SpotiPY, which allows you to access the Spotify data via Python console. My new idea is to create a track recommendation algorith. Basically the idea is to find tracks that are similar in terms of available audio features like danceability or instrumentality.

For now I would need your input to check and align the way the algorithm works.

I would like to ask you to mention one track you liked recently (one condition: it has to be a single from an album, the individually released singles wouldn't work here, genre doesn't matter).

As a response I will send you three other tracks from the same album the code identified as most similar.

Then I would like you to let me know what are the tracks ranked 2 to 4 in the same album to compare the results and adapt the code to match the best tracks better.

After I would get more data I would be happy to post the code and the input data somewhere so that you can check that and participate in developing it as well.
 
Hi Synonymous! Thank you for your response, especially as you are the first person. Below you can see the chart I prepared using my algorithm. On the X axis you can see all of the tracks from the Schoolboy Q - CrasH Talk album your song comes from, whereas Y axis contains the dissimilarity calculated by my script. The conclusions are that the most similar tracks to 'Numb Numb Juice' are:

- CrasH,
- Black Folk,
- Lies


Whereas Dangerous with Kid Cudi on the feat seems to be completely different in terms of instrumentality, danceability and some other features I took into account. For now I have an assumption that the more a track is different from the reference one you mentioned, the less you will like it. To complete please let me know which tracks from this album would you rank 2-4 as well as 'worst' 3. Explaining why you chose these tracks will be useful for me as well, as I would have some hints where I can improve my script.

Once again many thanks that you responded!

Synonymous.png
 
Hmmm interesting well I really like the tape so this is tough. Top 4 songs in my opinion is Crash, numb numb juice, Dangerous and black folk. Worst 3 would be Lies, Drunk and a tossup between chopstix and water. As for the best 4 songs I chose those because they have great bars, great production and it really has that Schoolboy q vibe where it really sounds like he wanted to do that record from his heart. They sounded authenitic I guess to me. The worst 3 doesnt mean they're bad but I guess you can say that they dont sound authentic in his character or his execution. It sounds more like he is using the vibe of the person featured and just trying to fit in whereas in the other records he is not tryna fit in hes just being natural and real. Idk if this makes sense and idk if you can even track that with that graph. But from that chart you gave I would say its pretty damn close to what i would expect so good stuff. Keep me updated.
 
Hmmm interesting well I really like the tape so this is tough. Top 4 songs in my opinion is Crash, numb numb juice, Dangerous and black folk. ...

Hi Synonymous and once again thank you for your post and thoughts, much appreciated! Interesting observation with tracks that remind you or not about the Schoolboy Q vibe, that's important notice for me, but makes my task difficult as style and its uniqueness cannot be represented with numbers:) Actually as for first try the prediction was not bad, so I'm happy with that. I only noticed that the 'worst' track on my side (with Kid Cudi) was your third choice. My idea for now is that I assumed that each audio feature (loudness, acousticness, danceability) is equally important to you and maybe a feature that is not important to you at all was totally different with the reference track. So in future I would add a request to mention the most and least important audio features for each participant and update the first post.
 
Oh yea lmao definitely that song was because of kid cudi but also because its a great song. Forgot to say that as a cudi fan Ill lean towards those songs. But yea just as you said taking features into the equation will help alot. I agree with the graph not being bad at all especially as the first go. But yes you're definitely right in everything youve said. Also, the song floating with 21 savage could also be in top 4 too. Feature was good and still felt like schoolboy q even though 21 savage was there. I liked the tape so it was hard for me to say top 4 etc.
 
Last edited:
Does it have to be released as a single too? Otherwise Lost Freestyle - Nas.

It doesn't matter actually - what I wanted was that you would provide ma with a song that was released within a longplay album, which you did and it is fine.

In your case it it a little bit different. My algorithm showed that two songs (You Mean The World To Me and Who Are You (ft. David Ranier)) are so different compared to your favorite track (their bar is so high) that the x-labels were cut (I will work on that and re-upload the chart as I will find out how to solve that). In your case there is a 'huge' (from the script point of view) loudness difference and this made those two tracks stand out from the others. Actually what do you think about these tracks?

I would predict your TOP3 (besides Lost Freestyle) to be: Highly Favored, Royalty (ft. RaVaughn) and Tanasia. Please let me know what do you think - your thoughts why would you choose tracks ranked 2-4 and worst 3 - would be really helpful and useful for me.

Thank you for participation!
 
Back
Top