{
  "paper": {
    "arxiv_id": "2606.25496",
    "title": "Recommendation as Generation: Unifying Personalized Video Generation and Recommendation at Industrial Scale",
    "url": "https://arxiv.org/abs/2606.25496",
    "organization": "Kuaishou / Beihang University"
  },
  "dataset": {
    "name": "MovieLens 100K",
    "users": 220,
    "items": 360
  },
  "setup": {
    "adapter": "rag-generation",
    "same_split_and_candidates": true
  },
  "baseline": {
    "name": "transition + content + popularity",
    "hit_at_10": 0.10909090909090909,
    "ndcg_at_10": 0.05400546778896325,
    "fresh_hit_at_10": 0.031746031746031744,
    "head_share_at_10": 0.23772727272727273
  },
  "method": {
    "name": "rag-generation core mechanism (validation blend=0.1)",
    "hit_at_10": 0.06818181818181818,
    "ndcg_at_10": 0.042345901609134776,
    "fresh_hit_at_10": 0.047619047619047616,
    "head_share_at_10": 0.14136363636363636
  },
  "relative": {
    "hit_at_10_percent": -37.5,
    "ndcg_at_10_percent": -21.589603159055066,
    "fresh_hit_at_10_percent": 50.0,
    "head_share_at_10_percent": -40.53537284894838
  },
  "stages": {
    "finite_scores": 360,
    "score_std": 0.6846373490293353,
    "sid_levels": 3,
    "sid_width": 8
  },
  "paper_results": {
    "revenue_vs_grm_lift_percent": 1.87,
    "revenue_vs_dlrm_lift_percent": 5.462
  },
  "scope": "执行论文可由公开 MovieLens 特征审计的核心计算；不复刻私有日志、生产大模型、多媒体生成器或线上服务栈。",
  "schema_version": 2,
  "manifest": {
    "adapter_key": "rag-generation",
    "arxiv_id": "2606.25496",
    "title": "Recommendation as Generation: Unifying Personalized Video Generation and Recommendation at Industrial Scale",
    "paper_url": "https://arxiv.org/abs/2606.25496",
    "track": "recommendation",
    "organization": "Kuaishou / Beihang University",
    "published": "2026-06-24",
    "code_url": null,
    "topics": [
      "generative-recommendation",
      "semantic-id",
      "video-generation"
    ],
    "local_code_dir": "src/auto_research/reproductions/industrial_2026",
    "fidelity": "core_mechanism",
    "evaluation_tier": "l2_public_dataset",
    "datasets": [
      "MovieLens 100K"
    ],
    "baseline": "fixed-catalog generative recommendation",
    "metrics": [
      "hit_at_10",
      "ndcg_at_10",
      "sid_levels",
      "sid_width"
    ],
    "default_seeds": [
      42,
      43,
      44
    ],
    "budget": "220 users / 360 items; validation-only blend selection",
    "device_capabilities": [
      "cpu",
      "mps",
      "cuda"
    ],
    "requires_gpu_validation": false,
    "gpu_validation_artifact": null,
    "online_evidence": [
      {
        "product": "Kuaishou advertising",
        "metric": "Ad Revenue vs GRM",
        "lift_percent": 1.87,
        "traffic": "large-scale production A/B test",
        "source_url": "https://arxiv.org/html/2606.25496v1",
        "source_location": "Section 4.1, Table 1",
        "retrieved_at": "2026-09-02"
      }
    ],
    "selection_exception": null,
    "evolve_operators": []
  },
  "provenance": {
    "created_at": "2026-09-01T16:17:30.716409+00:00",
    "code_commit": "1d956119a5ece6c757bd4acbc1ed0935420a7915",
    "python": "3.12.9",
    "platform": "sanitized public CPU run",
    "dataset_dir": "data",
    "dataset_fingerprint": "e66c261317a7b3179720ef3e3f5f79d7a204b135d1f217e2911b153ad2ce50a0",
    "packages": {
      "auto-research": "0.1.0",
      "numpy": "2.5.2",
      "torch": "2.13.0",
      "transformers": "5.14.1"
    },
    "artifact_path": "docs/reproductions/2606.25496-rag-generation/metrics/public-seeds42-44.json"
  },
  "evaluation_protocol": {
    "tier": "l2_public_dataset",
    "tier_label": "L2 公开数据集训练",
    "seeds": [
      42,
      43,
      44
    ],
    "budget": "220 users / 360 items; validation-only blend selection",
    "formal_comparison": true,
    "claim_policy": "formal multi-seed comparison"
  },
  "seed_results": [
    {
      "paper": {
        "arxiv_id": "2606.25496",
        "title": "Recommendation as Generation: Unifying Personalized Video Generation and Recommendation at Industrial Scale",
        "url": "https://arxiv.org/abs/2606.25496",
        "organization": "Kuaishou / Beihang University"
      },
      "dataset": {
        "name": "MovieLens 100K",
        "users": 220,
        "items": 360
      },
      "setup": {
        "adapter": "rag-generation",
        "same_split_and_candidates": true
      },
      "baseline": {
        "name": "transition + content + popularity",
        "hit_at_10": 0.10909090909090909,
        "ndcg_at_10": 0.05400546778896325,
        "fresh_hit_at_10": 0.031746031746031744,
        "head_share_at_10": 0.23772727272727273
      },
      "method": {
        "name": "rag-generation core mechanism (validation blend=0.1)",
        "hit_at_10": 0.06818181818181818,
        "ndcg_at_10": 0.042345901609134776,
        "fresh_hit_at_10": 0.047619047619047616,
        "head_share_at_10": 0.14136363636363636
      },
      "relative": {
        "hit_at_10_percent": -37.5,
        "ndcg_at_10_percent": -21.589603159055066,
        "fresh_hit_at_10_percent": 50.0,
        "head_share_at_10_percent": -40.53537284894838
      },
      "stages": {
        "finite_scores": 360,
        "score_std": 0.6846373490293353,
        "sid_levels": 3,
        "sid_width": 8
      },
      "paper_results": {
        "revenue_vs_grm_lift_percent": 1.87,
        "revenue_vs_dlrm_lift_percent": 5.462
      },
      "scope": "执行论文可由公开 MovieLens 特征审计的核心计算；不复刻私有日志、生产大模型、多媒体生成器或线上服务栈。"
    },
    {
      "paper": {
        "arxiv_id": "2606.25496",
        "title": "Recommendation as Generation: Unifying Personalized Video Generation and Recommendation at Industrial Scale",
        "url": "https://arxiv.org/abs/2606.25496",
        "organization": "Kuaishou / Beihang University"
      },
      "dataset": {
        "name": "MovieLens 100K",
        "users": 220,
        "items": 360
      },
      "setup": {
        "adapter": "rag-generation",
        "same_split_and_candidates": true
      },
      "baseline": {
        "name": "transition + content + popularity",
        "hit_at_10": 0.10909090909090909,
        "ndcg_at_10": 0.05400546778896325,
        "fresh_hit_at_10": 0.031746031746031744,
        "head_share_at_10": 0.23772727272727273
      },
      "method": {
        "name": "rag-generation core mechanism (validation blend=0.1)",
        "hit_at_10": 0.06818181818181818,
        "ndcg_at_10": 0.042345901609134776,
        "fresh_hit_at_10": 0.047619047619047616,
        "head_share_at_10": 0.14136363636363636
      },
      "relative": {
        "hit_at_10_percent": -37.5,
        "ndcg_at_10_percent": -21.589603159055066,
        "fresh_hit_at_10_percent": 50.0,
        "head_share_at_10_percent": -40.53537284894838
      },
      "stages": {
        "finite_scores": 360,
        "score_std": 0.6846373490293353,
        "sid_levels": 3,
        "sid_width": 8
      },
      "paper_results": {
        "revenue_vs_grm_lift_percent": 1.87,
        "revenue_vs_dlrm_lift_percent": 5.462
      },
      "scope": "执行论文可由公开 MovieLens 特征审计的核心计算；不复刻私有日志、生产大模型、多媒体生成器或线上服务栈。"
    },
    {
      "paper": {
        "arxiv_id": "2606.25496",
        "title": "Recommendation as Generation: Unifying Personalized Video Generation and Recommendation at Industrial Scale",
        "url": "https://arxiv.org/abs/2606.25496",
        "organization": "Kuaishou / Beihang University"
      },
      "dataset": {
        "name": "MovieLens 100K",
        "users": 220,
        "items": 360
      },
      "setup": {
        "adapter": "rag-generation",
        "same_split_and_candidates": true
      },
      "baseline": {
        "name": "transition + content + popularity",
        "hit_at_10": 0.10909090909090909,
        "ndcg_at_10": 0.05400546778896325,
        "fresh_hit_at_10": 0.031746031746031744,
        "head_share_at_10": 0.23772727272727273
      },
      "method": {
        "name": "rag-generation core mechanism (validation blend=0.1)",
        "hit_at_10": 0.06818181818181818,
        "ndcg_at_10": 0.042345901609134776,
        "fresh_hit_at_10": 0.047619047619047616,
        "head_share_at_10": 0.14136363636363636
      },
      "relative": {
        "hit_at_10_percent": -37.5,
        "ndcg_at_10_percent": -21.589603159055066,
        "fresh_hit_at_10_percent": 50.0,
        "head_share_at_10_percent": -40.53537284894838
      },
      "stages": {
        "finite_scores": 360,
        "score_std": 0.6846373490293353,
        "sid_levels": 3,
        "sid_width": 8
      },
      "paper_results": {
        "revenue_vs_grm_lift_percent": 1.87,
        "revenue_vs_dlrm_lift_percent": 5.462
      },
      "scope": "执行论文可由公开 MovieLens 特征审计的核心计算；不复刻私有日志、生产大模型、多媒体生成器或线上服务栈。"
    }
  ],
  "aggregate_metrics": {
    "dataset.users": {
      "mean": 220.0,
      "std": 0.0,
      "ci95": 0.0,
      "n": 3
    },
    "dataset.items": {
      "mean": 360.0,
      "std": 0.0,
      "ci95": 0.0,
      "n": 3
    },
    "baseline.hit_at_10": {
      "mean": 0.10909090909090909,
      "std": 0.0,
      "ci95": 0.0,
      "n": 3
    },
    "baseline.ndcg_at_10": {
      "mean": 0.05400546778896325,
      "std": 0.0,
      "ci95": 0.0,
      "n": 3
    },
    "baseline.fresh_hit_at_10": {
      "mean": 0.031746031746031744,
      "std": 0.0,
      "ci95": 0.0,
      "n": 3
    },
    "baseline.head_share_at_10": {
      "mean": 0.2377272727272727,
      "std": 0.0,
      "ci95": 0.0,
      "n": 3
    },
    "method.hit_at_10": {
      "mean": 0.06818181818181818,
      "std": 0.0,
      "ci95": 0.0,
      "n": 3
    },
    "method.ndcg_at_10": {
      "mean": 0.04234590160913478,
      "std": 0.0,
      "ci95": 0.0,
      "n": 3
    },
    "method.fresh_hit_at_10": {
      "mean": 0.047619047619047616,
      "std": 0.0,
      "ci95": 0.0,
      "n": 3
    },
    "method.head_share_at_10": {
      "mean": 0.14136363636363636,
      "std": 0.0,
      "ci95": 0.0,
      "n": 3
    },
    "relative.hit_at_10_percent": {
      "mean": -37.5,
      "std": 0.0,
      "ci95": 0.0,
      "n": 3
    },
    "relative.ndcg_at_10_percent": {
      "mean": -21.58960315905507,
      "std": 0.0,
      "ci95": 0.0,
      "n": 3
    },
    "relative.fresh_hit_at_10_percent": {
      "mean": 50.0,
      "std": 0.0,
      "ci95": 0.0,
      "n": 3
    },
    "relative.head_share_at_10_percent": {
      "mean": -40.53537284894838,
      "std": 0.0,
      "ci95": 0.0,
      "n": 3
    },
    "stages.finite_scores": {
      "mean": 360.0,
      "std": 0.0,
      "ci95": 0.0,
      "n": 3
    },
    "stages.score_std": {
      "mean": 0.6846373490293353,
      "std": 0.0,
      "ci95": 0.0,
      "n": 3
    },
    "stages.sid_levels": {
      "mean": 3.0,
      "std": 0.0,
      "ci95": 0.0,
      "n": 3
    },
    "stages.sid_width": {
      "mean": 8.0,
      "std": 0.0,
      "ci95": 0.0,
      "n": 3
    },
    "paper_results.revenue_vs_grm_lift_percent": {
      "mean": 1.87,
      "std": 0.0,
      "ci95": 0.0,
      "n": 3
    },
    "paper_results.revenue_vs_dlrm_lift_percent": {
      "mean": 5.462,
      "std": 0.0,
      "ci95": 0.0,
      "n": 3
    }
  },
  "manifest_ref": "reproduction:rag-generation"
}