fc / apps /documents /services /generation /litigation_generation_service.py
jun2024's picture
Upload 2241 files
b4f2238 verified
Raw History Blame Contribute Delete
14 kB
"""
诉讼文书生成服务
使用统一 LLM 服务生成诉讼文书(起诉状、答辩状).
主要改进(2026-01):
- 统一结构化输出解析
- 直接返回 Pydantic 对象
- 自动进行 schema 校验
Requirements: 3.1, 3.2, 3.3
"""
import logging
from datetime import date
from typing import Any, cast
from django.utils.translation import gettext_lazy as _
from apps.core.exceptions import NotFoundError, ValidationException
from apps.core.interfaces import ServiceLocator
from apps.core.utils.path import Path
from apps.documents.storage import get_docx_templates_root
from .litigation_context_builder import LitigationContextBuilder
from .litigation_llm_generator import LitigationLLMGenerator
from .outputs import ComplaintOutput, DefenseOutput
logger = logging.getLogger("apps.documents.generation")
class LitigationGenerationService:
"""
诉讼文书生成服务
使用提示词 + LLM + 结构化解析的生成链.
遵循四层架构规范:
- 依赖注入构造函数
- 延迟加载依赖
- 业务逻辑封装
"""
def _complaint_template_path(self) -> Path:
template_dir = get_docx_templates_root() / "2-案件材料"
return template_dir / "1-起诉材料" / "1-起诉状和反诉答辩状" / "1-起诉状.docx"
def _defense_template_path(self) -> Path:
template_dir = get_docx_templates_root() / "2-案件材料"
return template_dir / "2-答辩材料" / "1-答辩状和反诉状" / "1-答辩状.docx"
def __init__(
self,
llm_generator: LitigationLLMGenerator | None = None,
context_builder: LitigationContextBuilder | None = None,
) -> None:
"""
初始化服务(依赖注入)
Args:
llm_client: LLM 客户端实例(可选,延迟加载)
"""
self._llm_generator = llm_generator
self._context_builder = context_builder
@property
def llm_generator(self) -> LitigationLLMGenerator:
if self._llm_generator is None:
self._llm_generator = LitigationLLMGenerator()
return self._llm_generator
@property
def context_builder(self) -> LitigationContextBuilder:
if self._context_builder is None:
self._context_builder = LitigationContextBuilder()
return self._context_builder
def generate_complaint(self, case_data: dict[str, Any]) -> Any:
"""
生成起诉状
使用统一 LLM 结构化输出流程直接获取结构化结果.
Args:
case_data: 案件数据字典,包含以下字段:
- cause_of_action: 案由
- plaintiff: 原告
- defendant: 被告
- litigation_request: 诉讼请求
- facts_and_reasons: 事实与理由
Returns:
ComplaintOutput: 起诉状输出结构
Raises:
ValidationException: 生成失败或数据验证失败
Requirements: 3.1, 3.2, 3.3
"""
return self.llm_generator.generate_complaint(case_data)
def generate_defense(self, case_data: dict[str, Any]) -> Any:
"""
生成答辩状
使用统一 LLM 结构化输出流程直接获取结构化结果.
Args:
case_data: 案件数据字典,包含以下字段:
- cause_of_action: 案由
- plaintiff: 原告
- defendant: 被告
- defense_opinion: 答辩意见
- defense_reasons: 答辩理由
Returns:
DefenseOutput: 答辩状输出结构
Raises:
ValidationException: 生成失败或数据验证失败
Requirements: 3.1, 3.2, 3.3
"""
return self.llm_generator.generate_defense(case_data)
def generate_complaint_document(self, case_id: int, skip_llm: bool = True) -> tuple[str, bytes]:
"""
生成起诉状文档(返回文件名和字节流)
Args:
case_id: 案件 ID
skip_llm: 是否跳过 LLM 生成(默认 True,用于测试占位符)
Returns:
tuple[str, bytes]: (文件名, 文档字节流)
Raises:
NotFoundError: 案件不存在
ValidationException: 生成失败
"""
# 1. 获取案件数据
case_service = ServiceLocator.get_case_service()
case_dto = case_service.get_case_by_id_internal(case_id)
if not case_dto:
raise NotFoundError(message=_("案件不存在"), code="CASE_NOT_FOUND", errors={"case_id": case_id})
# 2. 提取案件信息
case_data = self.context_builder.extract_complaint_prompt_data(case_dto)
# 3. 使用 LLM 生成内容或使用模拟数据
if skip_llm:
logger.info("跳过 LLM 生成,使用模拟数据测试占位符")
llm_result = self._get_mock_complaint_output(case_data)
else:
llm_result = self.generate_complaint(case_data)
# 4. 准备模板上下文
context = self.context_builder.build_complaint_context(case_dto=case_dto, llm_result=llm_result)
# 5. 生成文件名
filename = self._generate_filename(case_id, "complaint")
logger.info("生成起诉状文件名: %s", filename)
# 6. 渲染模板
doc_bytes = self._render_template(self._complaint_template_path(), context)
return filename, doc_bytes
def generate_defense_document(self, case_id: int, skip_llm: bool = True) -> tuple[str, bytes]:
"""
生成答辩状文档(返回文件名和字节流)
Args:
case_id: 案件 ID
skip_llm: 是否跳过 LLM 生成(默认 True,用于测试占位符)
Returns:
tuple[str, bytes]: (文件名, 文档字节流)
Raises:
NotFoundError: 案件不存在
ValidationException: 生成失败
"""
# 1. 获取案件数据
case_service = ServiceLocator.get_case_service()
case_dto = case_service.get_case_by_id_internal(case_id)
if not case_dto:
raise NotFoundError(message=_("案件不存在"), code="CASE_NOT_FOUND", errors={"case_id": case_id})
# 2. 提取案件信息
case_data = self.context_builder.extract_defense_prompt_data(case_dto)
# 3. 使用 LLM 生成内容或使用模拟数据
if skip_llm:
logger.info("跳过 LLM 生成,使用模拟数据测试占位符")
llm_result = self._get_mock_defense_output(case_data)
else:
llm_result = self.generate_defense(case_data)
# 4. 准备模板上下文
context = self.context_builder.build_defense_context(case_dto=case_dto, llm_result=llm_result)
# 5. 生成文件名
filename = self._generate_filename(case_id, "defense")
logger.info("生成答辩状文件名: %s", filename)
# 6. 渲染模板
doc_bytes = self._render_template(self._defense_template_path(), context)
return filename, doc_bytes
def get_preview_context(self, case_id: int, litigation_type: str) -> dict[str, str]:
case_service = ServiceLocator.get_case_service()
case_dto = case_service.get_case_by_id_internal(case_id)
if not case_dto:
raise NotFoundError(message=_("案件不存在"), code="CASE_NOT_FOUND", errors={"case_id": case_id})
if litigation_type == "complaint":
template_path = self._complaint_template_path()
case_data = self.context_builder.extract_complaint_prompt_data(case_dto)
llm_result = self._get_mock_complaint_output(case_data)
context = self.context_builder.build_complaint_context(case_dto=case_dto, llm_result=llm_result)
elif litigation_type == "defense":
template_path = self._defense_template_path()
case_data = self.context_builder.extract_defense_prompt_data(case_dto)
llm_result = self._get_mock_defense_output(case_data)
context = self.context_builder.build_defense_context(case_dto=case_dto, llm_result=llm_result)
else:
raise ValidationException(
message=_("不支持的诉讼类型: %(t)s") % {"t": litigation_type},
code="INVALID_LITIGATION_TYPE",
)
from .pipeline import DocxPreviewService
preview_result = DocxPreviewService().preview(template_path, context)
return cast(dict[str, str], preview_result)
def _generate_filename(self, case_id: int, doc_type: str) -> str:
"""
生成文件名
Args:
case_id: 案件 ID
doc_type: 文档类型 ('complaint' 或 'defense')
Returns:
str: 生成的文件名
Requirements: 10.5
"""
from apps.documents.services.placeholders.litigation import FilenameService
filename_service = FilenameService()
if doc_type == "complaint":
return str(filename_service.generate_complaint_filename(case_id))
if doc_type == "defense":
return str(filename_service.generate_defense_filename(case_id))
raise ValidationException(
message=_("不支持的文档类型: %(t)s") % {"t": doc_type},
code="INVALID_DOC_TYPE",
errors={"doc_type": doc_type},
)
def _get_mock_complaint_output(self, case_data: dict[str, Any]) -> ComplaintOutput:
"""
获取模拟的起诉状输出(用于测试占位符)
Args:
case_data: 案件数据
Returns:
模拟的 ComplaintOutput
"""
from .outputs import PartyInfo
return ComplaintOutput(
title=f"{case_data.get('cause_of_action', '民事纠纷')}起诉状",
parties=[
PartyInfo(
name=case_data.get("plaintiff", "原告姓名"),
role="原告",
id_number="110101199001011234",
address="北京市朝阳区测试路1号",
),
PartyInfo(
name=case_data.get("defendant", "被告姓名"),
role="被告",
id_number="110101199002021234",
address="北京市海淀区测试路2号",
),
],
litigation_request="一、请求判令被告支付原告欠款人民币100,000元;\n二、请求判令被告承担本案诉讼费用.",
facts_and_reasons=(
"原告与被告于2023年1月1日签订借款合同,"
"约定被告向原告借款人民币100,000元,"
"借款期限为一年,到期后被告应当归还本金及利息.\n"
"借款到期后,原告多次催告被告还款,"
"但被告一直拖延不还.\n"
"综上所述,被告的行为已经严重侵害了原告的合法权益,"
"为维护原告的合法权益,"
"特向贵院提起诉讼,请求依法判决."
),
evidence=["借款合同", "转账记录", "催款记录"],
)
def _get_mock_defense_output(self, case_data: dict[str, Any]) -> DefenseOutput:
"""
获取模拟的答辩状输出(用于测试占位符)
Args:
case_data: 案件数据
Returns:
模拟的 DefenseOutput
"""
from .outputs import PartyInfo
return DefenseOutput(
title=f"{case_data.get('cause_of_action', '民事纠纷')}答辩状",
parties=[
PartyInfo(
name=case_data.get("plaintiff", "原告姓名"),
role="原告",
id_number="110101199001011234",
address="北京市朝阳区测试路1号",
),
PartyInfo(
name=case_data.get("defendant", "被告姓名"),
role="被告",
id_number="110101199002021234",
address="北京市海淀区测试路2号",
),
],
defense_opinion="答辩人不同意原告的全部诉讼请求",
defense_reasons=(
"一、原告所述借款事实不存在,"
"答辩人从未向原告借款.\n"
"二、原告提供的所谓借款合同系伪造,"
"答辩人从未在该合同上签字.\n"
"三、原告提供的转账记录与本案无关,"
"该笔款项系其他经济往来.\n"
"综上所述,原告的诉讼请求缺乏事实和法律依据,"
"请求贵院依法驳回原告的全部诉讼请求."
),
evidence=["银行流水", "聊天记录", "证人证言"],
)
def _render_template(self, template_path: Path, context: dict[str, Any]) -> bytes:
"""渲染 docx 模板"""
if not template_path.exists():
raise ValidationException(
message=_("模板文件不存在: %(p)s") % {"p": template_path},
code="TEMPLATE_NOT_FOUND",
errors={"template_path": str(template_path)},
)
try:
# 记录上下文键用于调试
logger.info("模板上下文包含的占位符: %s", list(context.keys()))
if "年份" not in context:
context["年份"] = str(date.today().year)
from .pipeline import DocxRenderer
rendered_content = DocxRenderer().render(str(template_path), context)
return cast(bytes, rendered_content)
except Exception as e:
logger.error("模板渲染失败: %s", e, exc_info=True)
raise ValidationException(
message=_("模板渲染失败: %(e)s") % {"e": e}, code="TEMPLATE_RENDER_ERROR", errors={"error": str(e)}
) from e