Hot Music Recommender

Kim Desi

Ironhack | May 2021

Content

• Project Description📙
• Questions❔
• Dataset📄
• Workflow🔧
• Organization📖
• Links💫

Project Description📙

The goal of this project was to create a music recommender based on the billboard hot 100 using a machine learning model. In the first step the music recommender should ask the user for a song title as input and then check whether the user input is in the current hot 100 or not.

• If it is, recommend another song that's currently in the top 100.
• If it's not, use a model trained on songs audio features to recommend a song that is similar to the user input

The recommendation is made by clustering the audio features of the dataset by similarity, using the K-means algorithm.

Questions❔

Starting this project I already had some questions, which only increased as soon as I started!

• How can you recommend music based on the user inputs song?
• Which audio features are logical to use in the recommendation model and lead to good performance?
• How can I improve a machine learning model like this, which approaches are there?

Dataset📄

To be able to check wheter the user input is part of the hot 100, a dataset of the current hot 100 song titles is needed. The data was obtained by web scraping the hot 100 website and formatting it into a dataframe. To be able to recommend songs based on audio features, a dataset of songs and their audio features is needed. In this case I used Spotify's API to obtain data about songs and their adio features and stored them into a dataframe.

Workflow🔧

1. Set up project plan

• think about what to do & research how to do it
• create different notebooks for different purposes
• set up folder structure in repository
• create OneNote for organization

2. Web scraping billboard hot 100

• scrape and format HTML of the website
• store data in dataframes & csv files

3. Spotify API

• reading documentation & setting up access
• obtain audio features of the hot 100
• obtain audio features of a playlist to create data for training the model
• store data in dataframes &csv files

4. Machine learning model

• prepare the data for the model to ensure optimal performance
• find optimal amount of clusters
• create new dataframes containing new information about which clusters songs belong to
• store data in dataframes/csv files and model using pickle

5. Hot music recommender

• create function including user input and recommendation
• test the recommender!

6. Revisit & improve

• revisit project and add information/comments in code
• improve code

Organization📖

This project was organized using a OneNote notebook.

Links💫

Repository