Analizing liked songs features from Spotify, using Bayesian networks

Modern streaming platforms are known for their ability to predict the preferences of their users. The music industry in particular poses a complex challenge due to the vastness of different music genres and songs available.
This project aims to model a Bayesian Network from a dataset built by fetching Spotify's API personal data and to experiment with different queries and methods in order to find interesting relationships. Finally, a use case scenario with the final model is presented.

• Report
• Powerpoint presentation

Screenshot

Relevant files

• Dataset.ipynb: notebook that generates the dataset, with all the explanations on how data was processed, and the reasons behind each choice.
• Bayesify.ipynb: main notebook where the models are built, and all the experiments performed.
• spotifyData.csv: the preprocessed dataset used for esimating the CPDs.

Libraries

• Spotipy to retrieve all kinds of data regarding my liked pieces, and converting it in csv to be imported by Pandas.
• Pgmpy to make Bayesian networks and inferences.
• Numpy, Seaborn, Matplotlib and Pandas for data manipulation and visualization.

References

1. Spotify's API
2. PGMpy Sampling source code
3. PGMpy rejection sampling source code