| <!doctype html> |
| <html lang="en"> |
| <head> |
| <meta charset="utf-8" /> |
| <meta name="viewport" content="width=device-width, initial-scale=1" /> |
| <title>NewsReX — Pre-trained News Recommendation Models</title> |
| <link rel="stylesheet" href="style.css" /> |
| </head> |
| <body> |
| <div class="container"> |
| <header> |
| <h1>NewsReX</h1> |
| <p class="subtitle">Pre-trained News Recommendation Models</p> |
| <div class="badges"> |
| <a href="https://arxiv.org/abs/2508.21572"><img src="https://img.shields.io/badge/arXiv-2508.21572-b31b1b.svg" alt="arXiv"></a> |
| <a href="https://github.com/igor17400/NewsReX"><img src="https://img.shields.io/badge/GitHub-NewsReX-blue.svg" alt="GitHub"></a> |
| <a href="https://www.python.org/downloads/release/python-3120/"><img src="https://img.shields.io/badge/python-3.12+-blue.svg" alt="Python 3.12+"></a> |
| </div> |
| </header> |
|
|
| <section> |
| <p>This organization hosts pre-trained weights for <strong>10 neural news recommendation models</strong> trained on the <a href="https://msnews.github.io/">MIND-small</a> dataset using the <a href="https://github.com/igor17400/NewsReX">NewsReX</a> framework. All models are trained with 3 random seeds (42, 123, 456) and evaluated on the standard MIND test split.</p> |
| </section> |
|
|
| <section> |
| <h2>Benchmark Results (MIND-small, mean ± std over 3 seeds)</h2> |
|
|
| <h3>JAX Models</h3> |
| <div class="table-wrapper"> |
| <table> |
| <thead> |
| <tr><th>Model</th><th>AUC</th><th>MRR</th><th>NDCG@5</th><th>NDCG@10</th><th>Weights</th></tr> |
| </thead> |
| <tbody> |
| <tr><td><strong>CROWN</strong></td><td>0.6778±0.0030</td><td>0.3246±0.0018</td><td>0.3619±0.0022</td><td>0.4233±0.0022</td><td><a href="https://huggingface.co/newsrex/CROWN-JAX-MIND-small">Download</a></td></tr> |
| <tr><td><strong>DIGAT</strong></td><td>0.6760±0.0021</td><td>0.3245±0.0021</td><td>0.3594±0.0035</td><td>0.4220±0.0027</td><td><a href="https://huggingface.co/newsrex/DIGAT-JAX-MIND-small">Download</a></td></tr> |
| <tr><td><strong>CAUM</strong></td><td>0.6734±0.0013</td><td>0.3202±0.0009</td><td>0.3546±0.0009</td><td>0.4185±0.0006</td><td><a href="https://huggingface.co/newsrex/CAUM-JAX-MIND-small">Download</a></td></tr> |
| <tr><td><strong>TCCM</strong></td><td>0.6734±0.0055</td><td>0.3208±0.0034</td><td>0.3574±0.0046</td><td>0.4194±0.0043</td><td><a href="https://huggingface.co/newsrex/TCCM-JAX-MIND-small">Download</a></td></tr> |
| <tr><td><strong>PP-Rec</strong></td><td>0.6676±0.0040</td><td>0.3182±0.0033</td><td>0.3544±0.0041</td><td>0.4164±0.0036</td><td><a href="https://huggingface.co/newsrex/PPREC-JAX-MIND-small">Download</a></td></tr> |
| <tr><td><strong>LSTUR</strong></td><td>0.6672±0.0020</td><td>0.3177±0.0033</td><td>0.3525±0.0037</td><td>0.4156±0.0033</td><td><a href="https://huggingface.co/newsrex/LSTUR-JAX-MIND-small">Download</a></td></tr> |
| <tr><td><strong>NAML</strong></td><td>0.6639±0.0014</td><td>0.3130±0.0022</td><td>0.3456±0.0033</td><td>0.4097±0.0025</td><td><a href="https://huggingface.co/newsrex/NAML-JAX-MIND-small">Download</a></td></tr> |
| <tr><td><strong>GLORY</strong></td><td>0.6624±0.0030</td><td>0.3152±0.0038</td><td>0.3483±0.0041</td><td>0.4119±0.0040</td><td><a href="https://huggingface.co/newsrex/GLORY-JAX-MIND-small">Download</a></td></tr> |
| <tr><td><strong>MINER</strong></td><td>0.6579±0.0024</td><td>0.3117±0.0027</td><td>0.3444±0.0035</td><td>0.4080±0.0025</td><td><a href="https://huggingface.co/newsrex/MINER-JAX-MIND-small">Download</a></td></tr> |
| <tr><td><strong>NRMS</strong></td><td>0.6561±0.0006</td><td>0.3075±0.0008</td><td>0.3394±0.0003</td><td>0.4039±0.0007</td><td><a href="https://huggingface.co/newsrex/NRMS-JAX-MIND-small">Download</a></td></tr> |
| </tbody> |
| </table> |
| </div> |
|
|
| <h3>PyTorch Models</h3> |
| <div class="table-wrapper"> |
| <table> |
| <thead> |
| <tr><th>Model</th><th>AUC</th><th>MRR</th><th>NDCG@5</th><th>NDCG@10</th><th>Weights</th></tr> |
| </thead> |
| <tbody> |
| <tr><td><strong>CROWN</strong></td><td>0.6705±0.0045</td><td>0.3183±0.0049</td><td>0.3553±0.0056</td><td>0.4173±0.0056</td><td><a href="https://huggingface.co/newsrex/CROWN-PYTORCH-MIND-small">Download</a></td></tr> |
| <tr><td><strong>CAUM</strong></td><td>0.6656±0.0053</td><td>0.3176±0.0028</td><td>0.3504±0.0040</td><td>0.4149±0.0035</td><td><a href="https://huggingface.co/newsrex/CAUM-PYTORCH-MIND-small">Download</a></td></tr> |
| <tr><td><strong>NAML</strong></td><td>0.6654±0.0015</td><td>0.3105±0.0009</td><td>0.3464±0.0027</td><td>0.4097±0.0018</td><td><a href="https://huggingface.co/newsrex/NAML-PYTORCH-MIND-small">Download</a></td></tr> |
| <tr><td><strong>PP-Rec</strong></td><td>0.6631±0.0044</td><td>0.3130±0.0024</td><td>0.3487±0.0041</td><td>0.4111±0.0033</td><td><a href="https://huggingface.co/newsrex/PPREC-PYTORCH-MIND-small">Download</a></td></tr> |
| <tr><td><strong>TCCM</strong></td><td>0.6616±0.0019</td><td>0.3088±0.0022</td><td>0.3428±0.0031</td><td>0.4057±0.0024</td><td><a href="https://huggingface.co/newsrex/TCCM-PYTORCH-MIND-small">Download</a></td></tr> |
| <tr><td><strong>NRMS</strong></td><td>0.6534±0.0025</td><td>0.3052±0.0021</td><td>0.3367±0.0019</td><td>0.4017±0.0022</td><td><a href="https://huggingface.co/newsrex/NRMS-PYTORCH-MIND-small">Download</a></td></tr> |
| <tr><td><strong>LSTUR</strong></td><td>—</td><td>—</td><td>—</td><td>—</td><td><a href="https://huggingface.co/newsrex/LSTUR-PYTORCH-MIND-small">Download</a></td></tr> |
| <tr><td><strong>DIGAT</strong></td><td>—</td><td>—</td><td>—</td><td>—</td><td><a href="https://huggingface.co/newsrex/DIGAT-PYTORCH-MIND-small">Download</a></td></tr> |
| <tr><td><strong>GLORY</strong></td><td>—</td><td>—</td><td>—</td><td>—</td><td><a href="https://huggingface.co/newsrex/GLORY-PYTORCH-MIND-small">Download</a></td></tr> |
| </tbody> |
| </table> |
| </div> |
| </section> |
|
|
| <section> |
| <h2>Supported Models</h2> |
| <div class="table-wrapper"> |
| <table> |
| <thead> |
| <tr><th>Model</th><th>Paper</th><th>Venue</th></tr> |
| </thead> |
| <tbody> |
| <tr><td>NRMS</td><td>Neural News Recommendation with Multi-Head Self-Attention</td><td>EMNLP 2019</td></tr> |
| <tr><td>NAML</td><td>Neural News Recommendation with Attentive Multi-View Learning</td><td>EMNLP 2019</td></tr> |
| <tr><td>LSTUR</td><td>Neural News Recommendation with Long- and Short-term User Representations</td><td>ACL 2019</td></tr> |
| <tr><td>CROWN</td><td>Intent Disentanglement and Feature Self-Supervision for News Recommendation</td><td>WWW 2025</td></tr> |
| <tr><td>PP-Rec</td><td>News Recommendation with Personalized User Interest and Popularity Deconfounding</td><td>ACL 2021</td></tr> |
| <tr><td>DIGAT</td><td>Dual Interactive Graph Attention Networks for News Recommendation</td><td>EMNLP 2022</td></tr> |
| <tr><td>GLORY</td><td>Global-Local News Recommendation via Multi-Channel Graph Modeling</td><td>NAACL 2024</td></tr> |
| <tr><td>MINER</td><td>Multi-Interest News Extraction and Recommendation</td><td>EMNLP 2022</td></tr> |
| <tr><td>CAUM</td><td>Candidate-Aware User Modeling for News Recommendation</td><td>RecSys 2023</td></tr> |
| <tr><td>TCCM</td><td>Topic-Centric Conversational Collaborative Model for News Recommendation</td><td>CIKM 2022</td></tr> |
| </tbody> |
| </table> |
| </div> |
| </section> |
|
|
| <section> |
| <h2>Quick Start</h2> |
| <pre><code>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</code></pre> |
| </section> |
|
|
| <section> |
| <h2>Repository Structure</h2> |
| <pre><code>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</code></pre> |
| </section> |
|
|
| <section> |
| <h2>Citation</h2> |
| <pre><code>@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}, |
| }</code></pre> |
| </section> |
|
|
| <footer> |
| <p><strong>Authors:</strong> Igor L.R. Azevedo (U. Tokyo) · Toyotaro Suzumura (U. Tokyo) · Yuichiro Yasui (Nikkei Inc.)</p> |
| </footer> |
| </div> |
| </body> |
| </html> |
|
|
