newsrex/CAUM-JAX-MIND-small-bert-random

NewsReX CAUM — JAX — MIND-small

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

Test Results

Seed	AUC	MRR	NDCG@5	NDCG@10
123	0.6764	0.3237	0.3593	0.4212
42 *	0.6825	0.3285	0.3631	0.4262
456	0.6802	0.3296	0.3621	0.4245
mean ± std	0.6797±0.0025	0.3273±0.0026	0.3615±0.0016	0.4240±0.0021

* Best seed (weights at repo root)

Experiment Configuration

model:
  name: caum
  architecture:
    news_encoder:
      type: multi_head_self_attention
      num_heads: 12
      head_dim: 20
      attention_hidden_dim: 200
      entity_embedding_dim: 100
      entity_num_heads: 4
      entity_head_dim: 40
      category_embedding_dim: 100
    user_encoder:
      type: candidate_aware
      candi_selfatt:
        num_heads: 20
        head_dim: 20
      candi_cnn:
        half_window: 1
      candi_att:
        hidden_dim: 400
        mid_dim: 256
    click_predictor:
      type: dot_product
  embedding:
    size: 300
    trainable: true
  news_dim: 400
  use_entity: true
  use_category: true
  dropout_rate: 0.2
  seed: 42
inputs:
  title:
    max_length: 30
  history:
    max_length: 50
  impressions:
    max_length: 5
  max_entities: 5
  process_title: true
  process_abstract: false
  process_category: true
  process_subcategory: false
  process_user_id: false
  process_entities: true
training:
  loss:
    name: categorical_crossentropy
    from_logits: true
    reduction: sum_over_batch_size
    label_smoothing: 0.0
  optimizer: adam
  learning_rate: 5.0e-05
  batch_size: 64
  num_epochs: 10
  gradient_clip_val: 1.0
  grad_accum_steps: 1
  early_stopping:
    patience: 3
    min_improvement: 0.01
  negative_sampling:
    strategy: random
    candidates: 4
evaluation:
  mode: fast
  evaluator: caum
  metrics:
  - auc
  - mrr
  - ndcg@5
  - ndcg@10
  batch_size: 256

Repository Structure

newsrex/CAUM-JAX-MIND-small-bert-random/
├── 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/glove/caum \
    framework=jax \
    weights=hf://newsrex/CAUM-JAX-MIND-small-bert-random/model.safetensors

# Run evaluation with a specific seed
uv run python src/eval.py \
    experiment=mind/glove/caum \
    framework=jax \
    weights=hf://newsrex/CAUM-JAX-MIND-small-bert-random/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},
}