Music to My Ear: Recommender System with Million Song Dataset

Xiaoyi Chen, Zhiran Chen, Kaicheng Ding, Weixin Liu, Xuening Wang, Ruitao Yi

Carniegie Mellon University

Introduction

We propose and implement a machine learning pipeline that combines content-based and collaborative recommendation methods for a large-scale, personalized song recommendation system. The goal is to predict which songs that a user will listen to and make a recommendation list of 10 songs to each user, given both the user’s listening history and full information (including meta-data and audio feature analysis) for all songs.

Dependencies

• Python 3.6
• Tables 3.6.1
• h5df 0.1.5
• Numpy 1.18
• Scikit-Learn 0.23.2
• Pandas 0.15.2
• Matplotlib 3.3.1
• Seaborn 0.10.1
• Spark_notebook_helpers 1.0.1

Files

.
├── utils
├── 10605_Project_Report.pdf
├── README.md
├── collaborative_bad_map.ipynb
├── collaborative_good_map.ipynb
├── dependencies.sh
├── preprocessing_zepplin.json
├── setup-script.sh
└── sid_mismatches.txt

Usages

Install Dependencies

To install dependencies, please run the following command to install everything required automatically:

$ ./setup-script.sh

Download Dataset

Download and extract dataset from here: http://millionsongdataset.com/ to designated directories.

Results

We achieved 0.4 recall when setting the cosine similarity threshold as 0.9 and 5.0524 RMSE with collabrative filtering

References

Fabio Aiolli. A preliminary study on a recommendersystem for the million songs dataset challenge. Volume964, 01, 2013.

Thierry Bertin-Mahieux, Daniel PW Ellis, Brian Whit-man, and Paul Lamere. The million song dataset. 2011.

Yi Li, Rudhir Gupta, Yoshiyuki Nagasaki, and TianheZhang. Million song dataset recommendation projectreport. 2012.

B. McFee, T. Bertin-Mahieux, D. Ellis, and G. Lanck-riet. The million song dataset challenge. 2012.