rebuild_payloads.py 5.3 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112
  1. """从 web/frameworks.json 重建 web/payloads.json(不调 LLM)——确定性。
  2. 顺带两个守卫修正:
  3. · 组件颗的 parent.how_id 必须指向同帖一个 how 颗,否则降级为 orphan 主颗(role=主, parent=null)。
  4. · 组件颗 payload 的「出自」用所属 how 的 title(而非裸 id)。
  5. 用法:python3 scripts/rebuild_payloads.py
  6. """
  7. from __future__ import annotations
  8. import json
  9. from pathlib import Path
  10. ROOT = Path(__file__).resolve().parent.parent
  11. FW = ROOT / "web/frameworks.json"
  12. PL = ROOT / "web/payloads.json"
  13. TYPE2ATTR = {"how": "how工序", "what": "what构成", "why": "why原理"}
  14. def _sections(blocks) -> list:
  15. out = []
  16. for b in blocks or []:
  17. head = b.get("小标题") or ""
  18. form = b.get("形式")
  19. out.append(f"【{head}】" + (f"({form})" if form else ""))
  20. if b.get("内容"):
  21. out.append(f" {b['内容']}")
  22. for it in b.get("条目") or []:
  23. word = it.get("词") or it.get("要素") or ""
  24. cue = it.get("选择线索")
  25. line = f" - {word}:{it.get('说明','')}" if word else f" - {it.get('说明','')}"
  26. if cue:
  27. line += f"(选用:{cue})"
  28. out.append(line)
  29. return out
  30. def content(k: dict):
  31. """How/Why → 拍平字符串;What → 结构化实体对象 {name,kind,维度拆分规则,body[]}。"""
  32. t = k["type"]
  33. if t == "how":
  34. L = [f"目标:{k.get('purpose','')}"]
  35. for i, s in enumerate(k.get("steps", []), 1):
  36. L += [f"步骤{i}", f" 输入:{s.get('input','')}",
  37. f" 方法:{s.get('directive','')}", f" 产出:{s.get('output','')}"]
  38. return "\n".join(L)
  39. if t == "what": # 结构化实体;概要走 custom_ext
  40. return {"name": k.get("title") or "", "kind": k.get("kind"),
  41. "维度拆分规则": k.get("维度拆分规则") if k.get("kind") == "子集" else None,
  42. "body": [{"item_name": it.get("item_name", ""), "item_desc": it.get("item_desc", ""),
  43. "作用域": it.get("作用域") or []} for it in (k.get("body") or [])]}
  44. return "\n".join([f"主张:{k.get('主张','')}"] + _sections(k.get("支撑")))
  45. def main() -> None:
  46. d = json.loads(FW.read_text(encoding="utf-8"))
  47. payloads = []
  48. demoted = 0
  49. for p in d["posts"]:
  50. ks = p["knowledges"]
  51. how_ids = {k["id"] for k in ks if k["type"] == "how"}
  52. how_titles = {k["id"]: k.get("title") for k in ks if k["type"] == "how"}
  53. for k in ks:
  54. k["业务阶段"] = [b for b in (k.get("业务阶段") or []) if b in ("灵感", "选题", "脚本")]
  55. for s in k.get("steps", []):
  56. if s.get("创作阶段") not in ("定向", "构思", "结构", "成文", "打磨"):
  57. s["创作阶段"] = None
  58. if k.get("role") == "组件" and (k.get("parent") or {}).get("how_id") not in how_ids:
  59. k["role"] = "主"
  60. k["parent"] = None
  61. demoted += 1
  62. for k in ks:
  63. t = k["type"]
  64. scopes, seen = [], set()
  65. def add(lst):
  66. for sc in lst:
  67. key = (sc["scope_type"], sc["value"])
  68. if key not in seen:
  69. seen.add(key)
  70. scopes.append({"scope_type": sc["scope_type"], "value": sc["value"]})
  71. if t == "how":
  72. for s in k.get("steps", []):
  73. add(s.get("作用域", []))
  74. else:
  75. add(k.get("作用域", []))
  76. ext = [{"key": "业务阶段", "type": "str", "value": v} for v in (k.get("业务阶段") or [])]
  77. if t == "what" and k.get("概要"):
  78. ext.append({"key": "概要", "type": "str", "value": k["概要"]})
  79. if t == "how":
  80. cs, cseen = [], set()
  81. for s in k.get("steps", []):
  82. c = s.get("创作阶段")
  83. if c and c not in cseen:
  84. cseen.add(c)
  85. cs.append(c)
  86. ext += [{"key": "创作阶段", "type": "str", "value": v} for v in cs]
  87. ext += [{"key": "动作", "type": "str", "value": s["动作"]} for s in k.get("steps", []) if s.get("动作")]
  88. if k.get("role") == "组件" and k.get("parent"):
  89. par = k["parent"]
  90. ext.append({"key": "出自", "type": "str",
  91. "value": f"{how_titles.get(par.get('how_id'), par.get('how_id'))} 第{par.get('step')}步"})
  92. payloads.append({
  93. "source": {"id": p.get("source_id") or f"xhs_{p['post_id']}", "source_type": "post", "title": p.get("title", ""),
  94. "author": "", "source_metadata": {"platform": p.get("platform"), "url": p.get("url")}},
  95. "title": k.get("title"), "content": content(k), "dim_creations": ["创作"],
  96. "dim_attributes": [TYPE2ATTR.get(t, "how工序")], "scopes": scopes, "custom_ext": ext})
  97. FW.write_text(json.dumps(d, ensure_ascii=False, indent=1), encoding="utf-8")
  98. PL.write_text(json.dumps(payloads, ensure_ascii=False, indent=2), encoding="utf-8")
  99. print(f"rebuilt {len(payloads)} payloads; 降级组件→orphan: {demoted}")
  100. if __name__ == "__main__":
  101. main()