This recommender uses user-based collaborative filtering.
The data (R2 - Yahoo! Music User Ratings of Songs with Artist, Album, and Genre Meta Information, v. 1.0) has been obtained from Yahoo! Webscope and consists of real data collected from Yahoo! Music services.
The New Data and New Challenges in Multimedia Research” by Bart Thomee, David A. Shamma, Gerald Friedland, Benjamin Elizalde, Karl Ni, Douglas Poland, Damian Borth, and Li-Jia Li, arXiv:1503.01817
In this project used are:
For model building:
train data
Sample
ind | user_id | song_id | rating |
---|---|---|---|
0 | 0 | 166 | 5 |
1 | 0 | 2245 | 4 |
2 | 0 | 3637 | 4 |
test data
Sample
ind | user_id | song_id | rating |
---|---|---|---|
0 | 0 | 7171 | 5 |
1 | 0 | 8637 | 4 |
2 | 0 | 21966 | 4 |
Metadata:
genres
Sample
genre_id | parent_genre_id | level | genre_name |
---|---|---|---|
0 | 0 | 1 | Unknown |
1 | 1 | 1 | Electronic/Dance |
2 | 1 | 2 | Ambient |
3 | 2 | 3 | Ambient Dub |
songs
Sample
song_id | album_id | artist_id | genre_id |
---|---|---|---|
0 | 12070 | 8490 | 0 |
1 | 19512 | 7975 | 134 |
2 | 18953 | 3492 | 0 |
1. Data preprocessing
2. Model training (SVD and kNN Baseline) and parameter tuning
3. Model performance visualization
4. Song recommender
5. Exploration on similiar users (SVD) and neighbours (kNN)
6. Music genres graph