{
  "paper": {
    "arxiv_id": "2502.16924",
    "title": "FilterLLM: Text-To-Distribution LLM for Billion-Scale Cold-Start Recommendation",
    "url": "https://arxiv.org/abs/2502.16924"
  },
  "dataset": {
    "name": "MovieLens-1M",
    "users": 420,
    "items": 640,
    "full_catalog_evaluation": true
  },
  "setup": {
    "seed": 42,
    "mechanism": "text_to_distribution"
  },
  "baseline": {
    "name": "content-to-item judgment",
    "hit_at_10": 0.14047619047619048,
    "ndcg_at_10": 0.07513562832974019,
    "mrr_at_10": 0.05541099773242629,
    "head_share_at_10": 0.21761904761904763
  },
  "method": {
    "name": "FilterLLM text-to-user-distribution + behavior guidance",
    "hit_at_10": 0.1261904761904762,
    "ndcg_at_10": 0.06775546607120683,
    "mrr_at_10": 0.049801587301587284,
    "head_share_at_10": 0.2714285714285714
  },
  "relative": {
    "hit_at_10_percent": -10.169491525423732,
    "ndcg_at_10_percent": -9.82245363829897,
    "mrr_at_10_percent": -10.12327996316948,
    "head_share_at_10_percent": 24.726477024070007
  },
  "diagnostics": {
    "distribution_rank": 24,
    "behavior_alignment": 0.7907814133209244,
    "validation_blend_weight": 0.75,
    "validation_ndcg_candidates": [
      0.059631,
      0.059816,
      0.06153,
      0.059298
    ]
  },
  "paper_results": {
    "cold_pv_percent": 5.13,
    "pctr_percent": 3.93,
    "gmv_percent": 10.86,
    "latency_reduction_percent": 97.12
  },
  "scope": "以 genre/title 特征映射用户分布，实际训练 ridge text-to-distribution、用户词表压缩及行为引导；没有复刻淘宝十亿用户词表、LLM 参数量和生产冷启动流量。",
  "schema_version": 2,
  "manifest_ref": "reproduction:filterllm",
  "evaluation_protocol": {
    "tier": "l2_public_dataset",
    "seeds": [
      42
    ],
    "formal_comparison": false,
    "claim_policy": "single/few-seed smoke result; do not claim a stable improvement"
  },
  "provenance": {
    "artifact_path": "docs/reproductions/2502.16924-filterllm/metrics/movielens-1m-seed42.json",
    "historical_migration": "historical-metrics-v2-2026-08-09",
    "original_code_commit": "not recorded",
    "dataset_fingerprint": "not recorded in historical artifact"
  }
}