jongpillee / deep-content-user

An implementation of "Deep Content-User Embedding Model for Music Recommendation" in Keras

deep-content-user

An implementation of "Deep Content-User Embedding Model for Music Recommendation, arxiv"

---

Requirements

• tensorflow-gpu==1.4.0
• keras==2.0.8

These requirements can be easily installed by: pip install -r requirements.txt

Scripts

• data_generator.py: The base script that contains batch data generator and valid set loader.
• load_label.py: The base script for metadata loading.
• mp3s_to_mel.py: Convert audio to mel-spectrogram.
• utils.py: Contain functions for evaluation.
• model.py: Basic and multi models.
• train.py: Module for training the recommendation models.
• encoding.py: Contains script for extracting embedding vector given the trained model weight.
• evaluation.py: embedding evaluation script for recommendation experiment.
• tagging.py: Module for training the tagging experiment.

Usage

Here are examples of how to run the code. (To run 1. and 2., you need MSD audio files and its related metadata from Echonest-TasteProfile-DataLoader, MSD_split)

1. python mp3s_to_mel.py
2. python train.py basic --margin 0.2 --N-negs 20
3. python encoding.py basic ./models/model_basic_20_0.20/weights.555-6.84.h5
4. python evaluation.py basic
5. python tagging.py basic

---

Reference

[1] Deep Content-User Embedding Model for Music Recommendation, Jongpil Lee, Kyungyun Lee, Jiyoung Park, Jangyeon Park, and Juhan Nam, arxiv, 2018