newsrex/CROWN-JAX-MIND-small

NewsReX CROWN — JAX — MIND-small

CROWN news recommendation model trained on MIND-small using the NewsReX framework (JAX).

Test Results

Seed	AUC	MRR	NDCG@5	NDCG@10
123	0.6742	0.3230	0.3605	0.4214
42 *	0.6816	0.3272	0.3650	0.4264
456	0.6777	0.3236	0.3602	0.4221
mean ± std	0.6778±0.0030	0.3246±0.0018	0.3619±0.0022	0.4233±0.0022

* Best seed (weights at repo root)

Experiment Configuration

model:
  name: crown
  architecture:
    news_encoder:
      type: transformer_intent_disentanglement
      num_heads: 10
      head_dim: 30
      feedforward_dim: 512
      num_layers: 1
      intent_num: 3
      intent_embedding_dim: 400
      attention_dim: 400
    user_encoder:
      type: bipartite_gnn_candidate_aware
      gnn_type: gat
      graph_num_layers: 5
      gat_num_heads: 4
      gat_alpha: 0.2
      user_attention_dim: 400
    click_predictor:
      type: dot_product
  embedding:
    size: 300
    trainable: true
  category_embedding_dim: 50
  subcategory_embedding_dim: 50
  dropout_rate: 0.2
  seed: 42
inputs:
  title:
    max_length: 32
  abstract:
    max_length: 128
  history:
    max_length: 50
  impressions:
    max_length: 5
  process_title: true
  process_abstract: true
  process_category: true
  process_subcategory: true
  process_user_id: false
training:
  loss:
    name: categorical_crossentropy
    from_logits: true
    reduction: sum_over_batch_size
    label_smoothing: 0.0
  auxiliary_alpha: 0.3
  optimizer: adam
  learning_rate: 0.0001
  batch_size: 64
  grad_accum_steps: 4
  num_epochs: 20
  gradient_clip_val: 4.0
  early_stopping:
    patience: 5
    min_improvement: 0.01
  negative_sampling:
    strategy: random
    candidates: 4
evaluation:
  mode: fast
  evaluator: default
  metrics:
  - auc
  - mrr
  - ndcg@5
  - ndcg@10
  batch_size: 64

Repository Structure

newsrex/CROWN-JAX-MIND-small/
├── model.safetensors          ← best seed (42)
├── test_results.json
├── training_run_summary.json
├── seed_123/model.safetensors
├── seed_42/model.safetensors
├── seed_456/model.safetensors
└── README.md

Usage

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

# Run evaluation with best seed weights
uv run python src/eval.py \
    experiment=mind/crown \
    framework=jax \
    weights=hf://newsrex/CROWN-JAX-MIND-small/model.safetensors

# Run evaluation with a specific seed
uv run python src/eval.py \
    experiment=mind/crown \
    framework=jax \
    weights=hf://newsrex/CROWN-JAX-MIND-small/seed_42/model.safetensors

Citation

@misc{newsrex2026,
  title={NewsReX: An Open-Source Multi-Framework for Neural News Recommendation},
  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},
}