{
  "paper": {
    "arxiv_id": "2606.03866",
    "title": "Taiji: Pareto Optimal Policy Optimization with Semantics-IDs Trade-off for Industrial LLM-Enhanced Recommendation",
    "url": "https://arxiv.org/abs/2606.03866",
    "organization": "Kuaishou Technology"
  },
  "dataset": {
    "name": "MovieLens 100K",
    "users": 220,
    "items": 360
  },
  "setup": {
    "adapter": "taiji",
    "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": "taiji core mechanism (validation blend=0.5)",
    "hit_at_10": 0.10454545454545454,
    "ndcg_at_10": 0.06447776884295099,
    "fresh_hit_at_10": 0.07936507936507936,
    "head_share_at_10": 0.1359090909090909
  },
  "relative": {
    "hit_at_10_percent": -4.166666666666668,
    "ndcg_at_10_percent": 19.39118663856458,
    "fresh_hit_at_10_percent": 150.0,
    "head_share_at_10_percent": -42.829827915869984
  },
  "stages": {
    "finite_scores": 360,
    "score_std": 0.20989833808284228,
    "pareto_objectives": 2,
    "adaptive_item_weights": 360
  },
  "paper_results": {
    "advertiser_value_lift_percent": 2.83,
    "revenue_lift_percent": 3.3
  },
  "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、生产大模型或线上服务栈。",
  "schema_version": 2,
  "manifest": {
    "adapter_key": "taiji",
    "arxiv_id": "2606.03866",
    "title": "Taiji: Pareto Optimal Policy Optimization with Semantics-IDs Trade-off for Industrial LLM-Enhanced Recommendation",
    "paper_url": "https://arxiv.org/abs/2606.03866",
    "track": "recommendation",
    "organization": "Kuaishou Technology",
    "published": "2026-06-02",
    "code_url": null,
    "topics": [
      "advertising",
      "reinforcement-alignment",
      "pareto-optimization"
    ],
    "local_code_dir": "src/auto_research/reproductions/taiji",
    "fidelity": "core_mechanism",
    "evaluation_tier": "l2_public_dataset",
    "datasets": [
      "MovieLens 100K"
    ],
    "baseline": "transition + content + popularity",
    "metrics": [
      "hit_at_10",
      "ndcg_at_10",
      "fresh_hit_at_10",
      "head_share_at_10"
    ],
    "default_seeds": [
      42,
      43,
      44
    ],
    "budget": "220 users / 360 items; validation-only blend selection",
    "device_capabilities": [
      "cpu"
    ],
    "requires_gpu_validation": false,
    "gpu_validation_artifact": null,
    "online_evidence": [
      {
        "product": "Kuaishou advertising",
        "metric": "Revenue",
        "lift_percent": 3.3,
        "traffic": "7 days; 10% control and 10% treatment",
        "source_url": "https://arxiv.org/html/2606.03866v1",
        "source_location": "Section 3.4, Table 2",
        "retrieved_at": "2026-09-02"
      }
    ],
    "selection_exception": null,
    "evolve_operators": [
      "reward:pareto-semantic-id"
    ]
  },
  "provenance": {
    "created_at": "2026-09-02T03:57:24.649850+00:00",
    "code_commit": "4786d139b957890340a7f58e54ee48d083b12944",
    "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/2606.03866-taiji/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": "paper-specific",
    "formal_comparison": true,
    "claim_policy": "formal multi-seed comparison"
  },
  "seed_results": [
    {
      "paper": {
        "arxiv_id": "2606.03866",
        "title": "Taiji: Pareto Optimal Policy Optimization with Semantics-IDs Trade-off for Industrial LLM-Enhanced Recommendation",
        "url": "https://arxiv.org/abs/2606.03866",
        "organization": "Kuaishou Technology"
      },
      "dataset": {
        "name": "MovieLens 100K",
        "users": 220,
        "items": 360
      },
      "setup": {
        "adapter": "taiji",
        "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": "taiji core mechanism (validation blend=0.5)",
        "hit_at_10": 0.10454545454545454,
        "ndcg_at_10": 0.06447776884295099,
        "fresh_hit_at_10": 0.07936507936507936,
        "head_share_at_10": 0.1359090909090909
      },
      "relative": {
        "hit_at_10_percent": -4.166666666666668,
        "ndcg_at_10_percent": 19.39118663856458,
        "fresh_hit_at_10_percent": 150.0,
        "head_share_at_10_percent": -42.829827915869984
      },
      "stages": {
        "finite_scores": 360,
        "score_std": 0.20989833808284228,
        "pareto_objectives": 2,
        "adaptive_item_weights": 360
      },
      "paper_results": {
        "advertiser_value_lift_percent": 2.83,
        "revenue_lift_percent": 3.3
      },
      "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、生产大模型或线上服务栈。",
      "runtime": {
        "requested_device": "auto",
        "cpu_threads": null,
        "platform": "Darwin arm64",
        "resolved_device": "cpu",
        "torch_version": "2.13.0",
        "accelerator": "arm"
      },
      "seed": 42,
      "schema_version": 2,
      "manifest": {
        "adapter_key": "taiji",
        "arxiv_id": "2606.03866",
        "title": "Taiji: Pareto Optimal Policy Optimization with Semantics-IDs Trade-off for Industrial LLM-Enhanced Recommendation",
        "paper_url": "https://arxiv.org/abs/2606.03866",
        "track": "recommendation",
        "organization": "Kuaishou Technology",
        "published": "2026-06-02",
        "code_url": null,
        "topics": [
          "advertising",
          "reinforcement-alignment",
          "pareto-optimization"
        ],
        "local_code_dir": "src/auto_research/reproductions/taiji",
        "fidelity": "core_mechanism",
        "evaluation_tier": "l2_public_dataset",
        "datasets": [
          "MovieLens 100K"
        ],
        "baseline": "transition + content + popularity",
        "metrics": [
          "hit_at_10",
          "ndcg_at_10",
          "fresh_hit_at_10",
          "head_share_at_10"
        ],
        "default_seeds": [
          42,
          43,
          44
        ],
        "budget": "220 users / 360 items; validation-only blend selection",
        "device_capabilities": [
          "cpu"
        ],
        "requires_gpu_validation": false,
        "gpu_validation_artifact": null,
        "online_evidence": [
          {
            "product": "Kuaishou advertising",
            "metric": "Revenue",
            "lift_percent": 3.3,
            "traffic": "7 days; 10% control and 10% treatment",
            "source_url": "https://arxiv.org/html/2606.03866v1",
            "source_location": "Section 3.4, Table 2",
            "retrieved_at": "2026-09-02"
          }
        ],
        "selection_exception": null,
        "evolve_operators": [
          "reward:pareto-semantic-id"
        ]
      },
      "provenance": {
        "created_at": "2026-09-02T03:57:24.649850+00:00",
        "code_commit": "4786d139b957890340a7f58e54ee48d083b12944",
        "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/2606.03866-taiji/metrics/public-seeds42-44.json"
      },
      "evaluation_protocol": {
        "tier": "l2_public_dataset",
        "tier_label": "L2 公开数据集训练",
        "seeds": [
          42
        ],
        "budget": "paper-specific",
        "formal_comparison": false,
        "claim_policy": "single/few-seed smoke result; do not claim a stable improvement"
      }
    },
    {
      "paper": {
        "arxiv_id": "2606.03866",
        "title": "Taiji: Pareto Optimal Policy Optimization with Semantics-IDs Trade-off for Industrial LLM-Enhanced Recommendation",
        "url": "https://arxiv.org/abs/2606.03866",
        "organization": "Kuaishou Technology"
      },
      "dataset": {
        "name": "MovieLens 100K",
        "users": 220,
        "items": 360
      },
      "setup": {
        "adapter": "taiji",
        "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": "taiji core mechanism (validation blend=0.5)",
        "hit_at_10": 0.10454545454545454,
        "ndcg_at_10": 0.06447776884295099,
        "fresh_hit_at_10": 0.07936507936507936,
        "head_share_at_10": 0.1359090909090909
      },
      "relative": {
        "hit_at_10_percent": -4.166666666666668,
        "ndcg_at_10_percent": 19.39118663856458,
        "fresh_hit_at_10_percent": 150.0,
        "head_share_at_10_percent": -42.829827915869984
      },
      "stages": {
        "finite_scores": 360,
        "score_std": 0.20989833808284228,
        "pareto_objectives": 2,
        "adaptive_item_weights": 360
      },
      "paper_results": {
        "advertiser_value_lift_percent": 2.83,
        "revenue_lift_percent": 3.3
      },
      "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、生产大模型或线上服务栈。",
      "runtime": {
        "requested_device": "auto",
        "cpu_threads": null,
        "platform": "Darwin arm64",
        "resolved_device": "cpu",
        "torch_version": "2.13.0",
        "accelerator": "arm"
      },
      "seed": 43,
      "schema_version": 2,
      "manifest": {
        "adapter_key": "taiji",
        "arxiv_id": "2606.03866",
        "title": "Taiji: Pareto Optimal Policy Optimization with Semantics-IDs Trade-off for Industrial LLM-Enhanced Recommendation",
        "paper_url": "https://arxiv.org/abs/2606.03866",
        "track": "recommendation",
        "organization": "Kuaishou Technology",
        "published": "2026-06-02",
        "code_url": null,
        "topics": [
          "advertising",
          "reinforcement-alignment",
          "pareto-optimization"
        ],
        "local_code_dir": "src/auto_research/reproductions/taiji",
        "fidelity": "core_mechanism",
        "evaluation_tier": "l2_public_dataset",
        "datasets": [
          "MovieLens 100K"
        ],
        "baseline": "transition + content + popularity",
        "metrics": [
          "hit_at_10",
          "ndcg_at_10",
          "fresh_hit_at_10",
          "head_share_at_10"
        ],
        "default_seeds": [
          42,
          43,
          44
        ],
        "budget": "220 users / 360 items; validation-only blend selection",
        "device_capabilities": [
          "cpu"
        ],
        "requires_gpu_validation": false,
        "gpu_validation_artifact": null,
        "online_evidence": [
          {
            "product": "Kuaishou advertising",
            "metric": "Revenue",
            "lift_percent": 3.3,
            "traffic": "7 days; 10% control and 10% treatment",
            "source_url": "https://arxiv.org/html/2606.03866v1",
            "source_location": "Section 3.4, Table 2",
            "retrieved_at": "2026-09-02"
          }
        ],
        "selection_exception": null,
        "evolve_operators": [
          "reward:pareto-semantic-id"
        ]
      },
      "provenance": {
        "created_at": "2026-09-02T03:57:25.160576+00:00",
        "code_commit": "4786d139b957890340a7f58e54ee48d083b12944",
        "python": "3.12.9",
        "platform": "macOS-26.5.2-arm64-arm-64bit",
        "dataset_dir": "/Users/bytedance/Documents/git_daiwk/auto-research/data",
        "dataset_fingerprint": "e66c261317a7b3179720ef3e3f5f79d7a204b135d1f217e2911b153ad2ce50a0",
        "packages": {
          "auto-research": "0.1.0",
          "numpy": "2.5.2",
          "torch": "2.13.0",
          "transformers": "5.14.1"
        }
      },
      "evaluation_protocol": {
        "tier": "l2_public_dataset",
        "tier_label": "L2 公开数据集训练",
        "seeds": [
          43
        ],
        "budget": "paper-specific",
        "formal_comparison": false,
        "claim_policy": "single/few-seed smoke result; do not claim a stable improvement"
      }
    },
    {
      "paper": {
        "arxiv_id": "2606.03866",
        "title": "Taiji: Pareto Optimal Policy Optimization with Semantics-IDs Trade-off for Industrial LLM-Enhanced Recommendation",
        "url": "https://arxiv.org/abs/2606.03866",
        "organization": "Kuaishou Technology"
      },
      "dataset": {
        "name": "MovieLens 100K",
        "users": 220,
        "items": 360
      },
      "setup": {
        "adapter": "taiji",
        "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": "taiji core mechanism (validation blend=0.5)",
        "hit_at_10": 0.10454545454545454,
        "ndcg_at_10": 0.06447776884295099,
        "fresh_hit_at_10": 0.07936507936507936,
        "head_share_at_10": 0.1359090909090909
      },
      "relative": {
        "hit_at_10_percent": -4.166666666666668,
        "ndcg_at_10_percent": 19.39118663856458,
        "fresh_hit_at_10_percent": 150.0,
        "head_share_at_10_percent": -42.829827915869984
      },
      "stages": {
        "finite_scores": 360,
        "score_std": 0.20989833808284228,
        "pareto_objectives": 2,
        "adaptive_item_weights": 360
      },
      "paper_results": {
        "advertiser_value_lift_percent": 2.83,
        "revenue_lift_percent": 3.3
      },
      "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、生产大模型或线上服务栈。",
      "runtime": {
        "requested_device": "auto",
        "cpu_threads": null,
        "platform": "Darwin arm64",
        "resolved_device": "cpu",
        "torch_version": "2.13.0",
        "accelerator": "arm"
      },
      "seed": 44,
      "schema_version": 2,
      "manifest": {
        "adapter_key": "taiji",
        "arxiv_id": "2606.03866",
        "title": "Taiji: Pareto Optimal Policy Optimization with Semantics-IDs Trade-off for Industrial LLM-Enhanced Recommendation",
        "paper_url": "https://arxiv.org/abs/2606.03866",
        "track": "recommendation",
        "organization": "Kuaishou Technology",
        "published": "2026-06-02",
        "code_url": null,
        "topics": [
          "advertising",
          "reinforcement-alignment",
          "pareto-optimization"
        ],
        "local_code_dir": "src/auto_research/reproductions/taiji",
        "fidelity": "core_mechanism",
        "evaluation_tier": "l2_public_dataset",
        "datasets": [
          "MovieLens 100K"
        ],
        "baseline": "transition + content + popularity",
        "metrics": [
          "hit_at_10",
          "ndcg_at_10",
          "fresh_hit_at_10",
          "head_share_at_10"
        ],
        "default_seeds": [
          42,
          43,
          44
        ],
        "budget": "220 users / 360 items; validation-only blend selection",
        "device_capabilities": [
          "cpu"
        ],
        "requires_gpu_validation": false,
        "gpu_validation_artifact": null,
        "online_evidence": [
          {
            "product": "Kuaishou advertising",
            "metric": "Revenue",
            "lift_percent": 3.3,
            "traffic": "7 days; 10% control and 10% treatment",
            "source_url": "https://arxiv.org/html/2606.03866v1",
            "source_location": "Section 3.4, Table 2",
            "retrieved_at": "2026-09-02"
          }
        ],
        "selection_exception": null,
        "evolve_operators": [
          "reward:pareto-semantic-id"
        ]
      },
      "provenance": {
        "created_at": "2026-09-02T03:57:25.506469+00:00",
        "code_commit": "4786d139b957890340a7f58e54ee48d083b12944",
        "python": "3.12.9",
        "platform": "macOS-26.5.2-arm64-arm-64bit",
        "dataset_dir": "/Users/bytedance/Documents/git_daiwk/auto-research/data",
        "dataset_fingerprint": "e66c261317a7b3179720ef3e3f5f79d7a204b135d1f217e2911b153ad2ce50a0",
        "packages": {
          "auto-research": "0.1.0",
          "numpy": "2.5.2",
          "torch": "2.13.0",
          "transformers": "5.14.1"
        }
      },
      "evaluation_protocol": {
        "tier": "l2_public_dataset",
        "tier_label": "L2 公开数据集训练",
        "seeds": [
          44
        ],
        "budget": "paper-specific",
        "formal_comparison": false,
        "claim_policy": "single/few-seed smoke result; do not claim a stable improvement"
      }
    }
  ],
  "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": 0.10454545454545454,
      "std": 0.0,
      "ci95": 0.0,
      "n": 3
    },
    "method.ndcg_at_10": {
      "mean": 0.06447776884295099,
      "std": 0.0,
      "ci95": 0.0,
      "n": 3
    },
    "method.fresh_hit_at_10": {
      "mean": 0.07936507936507936,
      "std": 0.0,
      "ci95": 0.0,
      "n": 3
    },
    "method.head_share_at_10": {
      "mean": 0.1359090909090909,
      "std": 0.0,
      "ci95": 0.0,
      "n": 3
    },
    "relative.hit_at_10_percent": {
      "mean": -4.166666666666668,
      "std": 0.0,
      "ci95": 0.0,
      "n": 3
    },
    "relative.ndcg_at_10_percent": {
      "mean": 19.39118663856458,
      "std": 0.0,
      "ci95": 0.0,
      "n": 3
    },
    "relative.fresh_hit_at_10_percent": {
      "mean": 150.0,
      "std": 0.0,
      "ci95": 0.0,
      "n": 3
    },
    "relative.head_share_at_10_percent": {
      "mean": -42.82982791586998,
      "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": 0.20989833808284228,
      "std": 0.0,
      "ci95": 0.0,
      "n": 3
    },
    "stages.pareto_objectives": {
      "mean": 2.0,
      "std": 0.0,
      "ci95": 0.0,
      "n": 3
    },
    "stages.adaptive_item_weights": {
      "mean": 360.0,
      "std": 0.0,
      "ci95": 0.0,
      "n": 3
    },
    "paper_results.advertiser_value_lift_percent": {
      "mean": 2.83,
      "std": 0.0,
      "ci95": 0.0,
      "n": 3
    },
    "paper_results.revenue_lift_percent": {
      "mean": 3.2999999999999994,
      "std": 0.0,
      "ci95": 0.0,
      "n": 3
    }
  },
  "manifest_ref": "reproduction:taiji"
}