NewsReX — Pre-trained News Recommendation Models

This organization hosts pre-trained weights for 10 neural news recommendation models trained on the MIND-small dataset using the NewsReX framework. All models are trained with 3 random seeds (42, 123, 456) and evaluated on the standard MIND test split.

Benchmark Results (MIND-small, mean ± std over 3 seeds)

JAX Models

Model	AUC	MRR	NDCG@5	NDCG@10	Weights
CROWN	0.6778±0.0030	0.3246±0.0018	0.3619±0.0022	0.4233±0.0022	Download
DIGAT	0.6760±0.0021	0.3245±0.0021	0.3594±0.0035	0.4220±0.0027	Download
CAUM	0.6734±0.0013	0.3202±0.0009	0.3546±0.0009	0.4185±0.0006	Download
TCCM	0.6734±0.0055	0.3208±0.0034	0.3574±0.0046	0.4194±0.0043	Download
PP-Rec	0.6676±0.0040	0.3182±0.0033	0.3544±0.0041	0.4164±0.0036	Download
LSTUR	0.6672±0.0020	0.3177±0.0033	0.3525±0.0037	0.4156±0.0033	Download
NAML	0.6639±0.0014	0.3130±0.0022	0.3456±0.0033	0.4097±0.0025	Download
GLORY	0.6624±0.0030	0.3152±0.0038	0.3483±0.0041	0.4119±0.0040	Download
MINER	0.6579±0.0024	0.3117±0.0027	0.3444±0.0035	0.4080±0.0025	Download
NRMS	0.6561±0.0006	0.3075±0.0008	0.3394±0.0003	0.4039±0.0007	Download

PyTorch Models

Model	AUC	MRR	NDCG@5	NDCG@10	Weights
CROWN	0.6705±0.0045	0.3183±0.0049	0.3553±0.0056	0.4173±0.0056	Download
CAUM	0.6656±0.0053	0.3176±0.0028	0.3504±0.0040	0.4149±0.0035	Download
NAML	0.6654±0.0015	0.3105±0.0009	0.3464±0.0027	0.4097±0.0018	Download
PP-Rec	0.6631±0.0044	0.3130±0.0024	0.3487±0.0041	0.4111±0.0033	Download
TCCM	0.6616±0.0019	0.3088±0.0022	0.3428±0.0031	0.4057±0.0024	Download
NRMS	0.6534±0.0025	0.3052±0.0021	0.3367±0.0019	0.4017±0.0022	Download
LSTUR	—	—	—	—	Download
DIGAT	—	—	—	—	Download
GLORY	—	—	—	—	Download

Supported Models

Model	Paper	Year
NRMS	Neural News Recommendation with Multi-Head Self-Attention	EMNLP 2019
NAML	Neural News Recommendation with Attentive Multi-View Learning	EMNLP 2019
LSTUR	Neural News Recommendation with Long- and Short-term User Representations	ACL 2019
CROWN	Intent Disentanglement and Feature Self-Supervision for News Recommendation	WWW 2025
PP-Rec	PP-Rec: News Recommendation with Personalized User Interest and Popularity Deconfounding	ACL 2021
DIGAT	Dual Interactive Graph Attention Networks for News Recommendation	EMNLP 2022
GLORY	Global-Local News Recommendation via Multi-Channel Graph Modeling	NAACL 2024
MINER	Multi-Interest News Extraction and Recommendation	EMNLP 2022
CAUM	Candidate-Aware User Modeling for News Recommendation	RecSys 2023
TCCM	Topic-Centric Conversational Collaborative Model for News Recommendation	CIKM 2022

Quick Start

git clone https://github.com/igor17400/NewsReX.git
cd NewsReX && uv sync --extra jax

# Evaluate a pre-trained model
uv run python src/train.py experiment=mind/nrms framework=jax \
    weights=hf://newsrex/NRMS-JAX-MIND-small/model.safetensors

# Train from scratch (3 seeds)
uv run python src/train.py experiment=mind/nrms framework=jax \
    multi_seed.enabled=true

Repository Structure

Each model repo follows this layout:

newsrex/{MODEL}-{FRAMEWORK}-MIND-small/
├── model.safetensors              <- best seed (default download)
├── test_results.json
├── training_run_summary.json
├── seed_42/model.safetensors
├── seed_123/model.safetensors
├── seed_456/model.safetensors
└── README.md

Citation

@misc{azevedo2025newsrex,
  title={NewsReX: A More Efficient Approach to News Recommendation with Keras 3 and JAX},
  author={Igor L. R. Azevedo and Toyotaro Suzumura and Yuichiro Yasui},
  year={2025},
  eprint={2508.21572},
  archivePrefix={arXiv},
  primaryClass={cs.IR},
  url={https://arxiv.org/abs/2508.21572},
}

Authors

• Igor L.R. Azevedo — The University of Tokyo
• Toyotaro Suzumura — The University of Tokyo
• Yuichiro Yasui — Nikkei Inc.