| 123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124 |
- """步骤2 结构化抽取 schema 定义:字段列表、Function Calling schema、提示词构造"""
- from __future__ import annotations
- from typing import Any
- # 抽取的固定字段列表(顺序即重要性顺序,limitations 为重点字段)
- EXTRACT_FIELDS: list[str] = [
- "title",
- "task_problem",
- "method",
- "key_contributions",
- "datasets",
- "metrics",
- "results",
- "limitations",
- "future_work",
- ]
- # 各字段中文描述,用于 schema 与提示词
- FIELD_DESCRIPTIONS: dict[str, str] = {
- "title": "论文标题",
- "task_problem": "研究任务/问题",
- "method": "核心方法",
- "key_contributions": "主要贡献",
- "datasets": "使用的数据集",
- "metrics": "评估指标",
- "results": "实验结果",
- "limitations": "局限性(重点字段,须论文明确承认或可从结果推断,严禁臆造)",
- "future_work": "未来工作",
- }
- def _field_source_schema(description: str) -> dict[str, Any]:
- """构造单个 FieldSource 字段的 JSON schema"""
- return {
- "type": "object",
- "description": description,
- "properties": {
- "value": {
- "type": "string",
- "description": "字段值",
- },
- "source_sections": {
- "type": "array",
- "items": {"type": "string"},
- "description": "原文出处章节标题列表,用于回溯核查",
- },
- "confidence": {
- "type": "number",
- "description": "置信度 0.0-1.0,信息明确且原文直述取高值,推断取中值,不足取低值",
- },
- },
- "required": ["value", "source_sections", "confidence"],
- }
- # DeepSeek function calling schema:强制模型输出上述字段的 JSON
- EXTRACT_FUNCTION_SCHEMA: dict[str, Any] = {
- "type": "function",
- "function": {
- "name": "extract_paper_fields",
- "description": (
- "从论文中抽取结构化字段,每个字段含 value/source_sections/confidence。"
- "必须返回所有字段,limitations 须真实可溯源。"
- ),
- "parameters": {
- "type": "object",
- "properties": {
- field: _field_source_schema(desc)
- for field, desc in FIELD_DESCRIPTIONS.items()
- },
- "required": list(EXTRACT_FIELDS),
- },
- },
- }
- def build_extraction_prompt(parsed_paper: dict) -> str:
- """构造抽取提示词(中文)
- 强调三点:
- 1. source_sections 必须标注原文出处章节;
- 2. limitations 须真实,论文未明确承认时不得臆造;
- 3. 严格输出 JSON,不添加额外说明。
- """
- title = parsed_paper.get("title", "")
- sections = parsed_paper.get("sections", [])
- # 拼接章节正文,保留 heading 以便 source_sections 标注
- sections_text = ""
- for i, sec in enumerate(sections, 1):
- heading = sec.get("heading", f"章节{i}")
- content = sec.get("content", "")
- captions = sec.get("figure_captions", []) or []
- sections_text += f"\n\n## {heading}\n{content}"
- if captions:
- sections_text += "\n\n图表标题:\n" + "\n".join(f"- {c}" for c in captions)
- fields_list = "、".join(EXTRACT_FIELDS)
- return f"""请从以下论文中抽取结构化字段。
- 论文标题:{title}
- 论文正文:
- {sections_text}
- 抽取要求:
- 1. 必须抽取以下字段:{fields_list}
- 2. 每个字段必须是 FieldSource 结构:
- {{
- "value": "字段值(字符串,尽量简洁完整)",
- "source_sections": ["原文出处章节标题列表"],
- "confidence": 0.0-1.0
- }}
- 3. source_sections 必须标注字段信息来源于上面哪个章节(用章节标题原文),便于回溯核查;可标注多个章节。
- 4. limitations 字段特别要求:必须是论文明确承认的局限,或可从实验结果直接推断的局限,严禁臆造或泛泛而谈。若论文未明确提及局限,value 填写"论文未明确陈述局限",confidence 设为较低值(如 0.3)。
- 5. confidence 反映你对该字段抽取结果的可信度:
- - 信息明确且原文直述 → 0.9-1.0
- - 需要从原文推断 → 0.6-0.8
- - 信息不足或模糊 → 低于 0.6
- 6. 严格输出 JSON,不要添加任何额外说明文字或 markdown 代码块标记。
- """
|