![]() ![]() The evaluation metrics used to judge each model include Precision, Recall and Catalog Coverage. Our Factorization Machines model was implemented using LightFM ( ). Our ALS model was implemented through Ben Frederickson's implicit package ( ). We also established a baseline model where the same top-k most popular artists (by total plays) are recommended to each user. The nature of Factorization Machines allowed us to add in user features in the form of the metadata we possess. We implemented and evaluated two main models to recommend artists to Last.fm users, one driven by Alternating Least Squares (ALS) and the other by Factorization Machines. Where possible, we utilized this information as additional features for our more advanced recommender system. We also worked with metadata that contained contextual information for each user, namely age, gender and country. Visualizations generated from our findings can be viewed from the 'Plots' folder.Īfter model selection, we then proceeded to expand our model to the full dataset and evaluated how it scaled and performed. This dataset and all relevant jupyter notebooks can be found in the 'Code' folder, with Main.ipynb being our primary markdown file. You can download the full dataset (~1.64 GB) here:įor this project, we started with a subset of the dataset that contained a total of 9,000 users and 47,102 artists. Model can be generalizable into a more production ready package in the future. Usage is currently only within the Jupyter Notebooks. This project aims to create a personalization/recommender system using the Last.fm Music dataset ( ). Last.Fm Music Personalization and Recommmender System
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |