This repository contains Priyam's music-related projects.
First, a one/some-vs-many Kpop artist classifier/recommendation system. The API loader contains a script (calling upon functions in Helpers) to compose a CSV file with Spotify artist, album, and song metadata as specified in Playlist. Generic Classifier takes two CSV files, the target (Mamamoo, EXID, and/or Red Velvet) and others (up to 100 kpop artists). Assigning a feature of 1 to the target and 0 to others, the tree-based classifier learns characteristics of the target. False positives indicate songs by different artists that have similar metadata features to the target artist. From a subjective perspective, the model works because the similar metadata correlates to a similar musical aesthetic. Generic Classifier, therefore, serves as the basis for Generic Recommender.
Generic Recommender simply takes one step further. First, I train the model on the target and others, as before. Then a new CSV file, containing relatively new artists, is processed the same way. The model then predicts which songs from these rookie artists resemble the target artist. This will make it possible to infer whether the rookie artists sufficiently resemble the target. The goal is to determine which kpop groups will succeed my top 3, and I can use this model on each of the 3 or on the composite.