{
  "paper": {
    "arxiv_id": "2502.00321",
    "title": "MIM: Multi-modal Content Interest Modeling Paradigm for User Behavior Modeling",
    "url": "https://arxiv.org/abs/2502.00321"
  },
  "dataset": {
    "name": "MovieLens-1M",
    "users": 420,
    "items": 640,
    "full_catalog_evaluation": true
  },
  "setup": {
    "seed": 42,
    "mechanism": "masked_multimodal_interest"
  },
  "baseline": {
    "name": "ID co-occurrence interest model",
    "hit_at_10": 0.14047619047619048,
    "ndcg_at_10": 0.07513562832974019,
    "mrr_at_10": 0.05541099773242629,
    "head_share_at_10": 0.21761904761904763
  },
  "method": {
    "name": "MIM masked multimodal pretraining + C-SFT + CiUBM",
    "hit_at_10": 0.12142857142857143,
    "ndcg_at_10": 0.07584470393503259,
    "mrr_at_10": 0.06228080120937263,
    "head_share_at_10": 0.3904761904761905
  },
  "relative": {
    "hit_at_10_percent": -13.559322033898306,
    "ndcg_at_10_percent": 0.9437275245514073,
    "mrr_at_10_percent": 12.39790611625489,
    "head_share_at_10_percent": 79.43107221006564
  },
  "diagnostics": {
    "masked_reconstruction_mse": 0.018867625416619935,
    "validation_blend_weight": 1.0,
    "validation_ndcg_candidates": [
      0.059486,
      0.068701,
      0.076633,
      0.077251
    ]
  },
  "paper_results": {
    "ctr_percent": 14.14,
    "rpm_percent": 4.12
  },
  "scope": "以 MovieLens genre 特征代理图文模态，实际执行遮盖重建、内容/协同对比对齐和内容兴趣融合；未复刻淘宝私有多模态编码器、曝光日志和线上 CiUBM serving。",
  "schema_version": 2,
  "manifest_ref": "reproduction:mim",
  "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/2502.00321-mim/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"
  }
}