Skip to content

deezer/recsys25-reacta

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

5 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

"Beyond the past": Leveraging Audio and Human Memory for Sequential Music Recommendation

This repository provides Python code to reproduce experiments from our paper:

Viet-Anh Tran, Bruno Sguerra, Gabriel Meseguer-Brocal, Lea Briand and Manuel Moussallam. "Beyond the past": Leveraging Audio and Human Memory for Sequential Music Recommendation. In: Proceedings of the 19th ACM Conference on Recommender Systems (RecSys 2025), September 2025.

Environment

  • python 3.9.13
  • tensorflow 2.11.0
  • tqdm 4.65.0
  • numpy 1.24.2
  • scipy 1.10.1
  • pandas 1.5.3
  • toolz 0.12.0

Dataset

The original anonymized version of our Deezer proprietary dataset (before filters applied in this work) can be freely downloaded from Zenodo. This dataset comprises nearly 900 million organically selected, time-stamped listening events from 4 million anonymized Deezer users recorded in 2023. It covers 50,000 anonymized songs, among the platform’s most popular, along with their multimodal pre-trained embedding vectors (Audio and SVD) generated by our internal model. All files are provided in Parquet format, readable with the pandas.read_parquet function.

Hyperparameters

Hperparameters for each model are found in the corresponding configuration file in configs directory:

  • Number of epochs: 100
  • Optimizer: Adam
  • Batch size: 512
  • Embedding dimension $d$: 128
  • $\alpha = 0.5$ for the Base-Level (BL) module in all ACT-R models.
  • For Transformer-based models (PISA, REACTA): $B=2, H=2$, and $L=30$.
  • Other hyperparameters were tuned via grid search on validation set:
    • Learning rates: {0.0002, 0.0005, 0.00075, 0.001}
    • $\lambda$: {0.0, 0.3, 0.5, 0.8, 0.9, 1.0}
    • $\beta$ and $\gamma$: {0.2, 0.4, 0.6, 0.8, 1.0}

About

"Beyond the past": Leveraging Audio and Human Memory for Sequential Music Recommendation

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published