This repo contains notebooks for building a recommender system for influencers and artists/tracks on Groover dataset.

• 01_exploration_and_preprocessing.ipynb provides quick exploration and preprocessing of categorical data

• 02_first_models.ipynb builds benchmark with random, naive and LGBM models, and first Neural Networks using Keras

• 03_final_models.ipynb builds Neural Networks models incorporating textual data from track_info

NB: plotly plots do not show on github, please run notebooks to display them.

1. install requirements.txt in a virtualenv: pip install -r requirements.txt
2. create ./data/raw/ and ./data/preprocessed/ directories
3. put raw data in ./data/raw/
4. install graphviz: sudo apt-get install graphviz
5. run 01_exploration_and_preprocessing.ipynb for preprocessing (preprocessed data is created and put in ./data/preprocessed/ directory)
6. run 02_first_models.ipynb to train models and benchmark
7. download and unzip cc.fr.300.vec french pre-trained word embeddings from this link and put it in ./
8. run 03_final_models.ipynb to train final models