445 lines
16 KiB
Python
445 lines
16 KiB
Python
|
"""
|
|||
|
|
事实密度 + 结构化块增强模块
|
|||
|
|
提升内容的事实信息密度和结构化程度
|
|||
|
|
"""
|
|||
|
|
from typing import Dict, List, Optional, Set
|
|||
|
|
from langchain_core.prompts import PromptTemplate
|
|||
|
|
from langchain_core.output_parsers import StrOutputParser
|
|||
|
|
import json
|
|||
|
|
import re
|
|||
|
|
|
|||
|
|
|
|||
|
|
class FactDensityEnhancer:
|
|||
|
|
"""事实密度和结构化块增强器"""
|
|||
|
|
|
|||
|
|
def __init__(self):
|
|||
|
|
# 事实密度和结构化评估 Prompt
|
|||
|
|
self.assessment_prompt_template = """
|
|||
|
|
你是内容质量评估专家，专门评估内容的事实密度和结构化程度。
|
|||
|
|
|
|||
|
|
【内容】
|
|||
|
|
{content}
|
|||
|
|
|
|||
|
|
【品牌】{brand}
|
|||
|
|
【优势】{advantages}
|
|||
|
|
【平台】{platform}
|
|||
|
|
|
|||
|
|
【评估标准】
|
|||
|
|
|
|||
|
|
1. **事实密度**（50分）
|
|||
|
|
评估内容中包含的事实性信息：
|
|||
|
|
- 数据信息：具体数字、百分比、统计数据（如"80%的用户"、"2024年数据显示"）
|
|||
|
|
- 案例信息：具体案例、实例、应用场景（如"某企业案例"、"实际应用表明"）
|
|||
|
|
- 标准信息：行业标准、规范、要求（如"ISO标准"、"行业规范"）
|
|||
|
|
- 对比信息：对比数据、差异说明（如"相比传统方案提升30%"）
|
|||
|
|
- 时间信息：时间节点、时效性（如"2024年"、"最新版本"）
|
|||
|
|
- 来源信息：数据来源、案例来源（如"根据XX报告"、"参考XX研究"）
|
|||
|
|
|
|||
|
|
2. **结构化块**（50分）
|
|||
|
|
评估内容的结构化元素：
|
|||
|
|
- 标题层级：是否有清晰的标题层级（H1/H2/H3）
|
|||
|
|
- 结论摘要：是否有开头的结论摘要（80-120字）
|
|||
|
|
- 清单列表：是否有清单、列表、要点（- 或 1. 格式）
|
|||
|
|
- FAQ部分：是否有常见问题解答
|
|||
|
|
- 代码块：技术内容是否有代码示例（如适用）
|
|||
|
|
- 对比表格：是否有对比表格或对比列表
|
|||
|
|
- 步骤说明：是否有步骤、流程说明
|
|||
|
|
- 总结部分：是否有结尾总结
|
|||
|
|
|
|||
|
|
【输出格式】
|
|||
|
|
请严格按照以下 JSON 格式输出，不要添加任何其他内容：
|
|||
|
|
|
|||
|
|
{{
|
|||
|
|
"scores": {{
|
|||
|
|
"fact_density": <事实密度得分 0-50>,
|
|||
|
|
"structure": <结构化得分 0-50>,
|
|||
|
|
"total": <总分 0-100>
|
|||
|
|
}},
|
|||
|
|
"fact_analysis": {{
|
|||
|
|
"data_count": <数据信息数量>,
|
|||
|
|
"case_count": <案例信息数量>,
|
|||
|
|
"standard_count": <标准信息数量>,
|
|||
|
|
"comparison_count": <对比信息数量>,
|
|||
|
|
"time_count": <时间信息数量>,
|
|||
|
|
"source_count": <来源信息数量>,
|
|||
|
|
"missing_facts": [
|
|||
|
|
"<缺失的事实类型1>",
|
|||
|
|
"<缺失的事实类型2>"
|
|||
|
|
]
|
|||
|
|
}},
|
|||
|
|
"structure_analysis": {{
|
|||
|
|
"has_title": <是否有标题层级 true/false>,
|
|||
|
|
"has_summary": <是否有结论摘要 true/false>,
|
|||
|
|
"has_list": <是否有清单列表 true/false>,
|
|||
|
|
"has_faq": <是否有FAQ true/false>,
|
|||
|
|
"has_code": <是否有代码块 true/false>,
|
|||
|
|
"has_table": <是否有对比表格 true/false>,
|
|||
|
|
"has_steps": <是否有步骤说明 true/false>,
|
|||
|
|
"has_conclusion": <是否有总结 true/false>,
|
|||
|
|
"missing_blocks": [
|
|||
|
|
"<缺失的结构化块1>",
|
|||
|
|
"<缺失的结构化块2>"
|
|||
|
|
]
|
|||
|
|
}},
|
|||
|
|
"details": {{
|
|||
|
|
"fact_density": "<事实密度评估详情>",
|
|||
|
|
"structure": "<结构化评估详情>"
|
|||
|
|
}},
|
|||
|
|
"improvements": [
|
|||
|
|
"<改进建议1>",
|
|||
|
|
"<改进建议2>",
|
|||
|
|
"<改进建议3>"
|
|||
|
|
]
|
|||
|
|
}}
|
|||
|
|
|
|||
|
|
【开始评估】
|
|||
|
|
"""
|
|||
|
|
|
|||
|
|
# 事实密度和结构化强化 Prompt
|
|||
|
|
self.enhancement_prompt_template = """
|
|||
|
|
你是内容优化专家，专门提升内容的事实密度和结构化程度。
|
|||
|
|
|
|||
|
|
【原内容】
|
|||
|
|
{content}
|
|||
|
|
|
|||
|
|
【品牌】{brand}
|
|||
|
|
【优势】{advantages}
|
|||
|
|
【平台】{platform}
|
|||
|
|
|
|||
|
|
【强化要求】
|
|||
|
|
|
|||
|
|
1. **事实密度强化**
|
|||
|
|
- 添加数据信息：在合适位置添加数据占位（如"根据XX数据显示，约XX%的企业"）
|
|||
|
|
- 添加案例信息：添加实际案例或应用场景（用占位符，如"某企业案例表明"）
|
|||
|
|
- 添加标准信息：提及相关标准或规范（如"按照XX标准"、"参考XX规范"）
|
|||
|
|
- 添加对比信息：添加对比数据或差异说明（如"相比传统方案，提升约XX%"）
|
|||
|
|
- 添加时间信息：明确时间节点或时效性（如"2024年最新"、"当前版本"）
|
|||
|
|
- 添加来源信息：标注数据来源（如"根据XX行业报告"、"参考XX研究"）
|
|||
|
|
|
|||
|
|
2. **结构化块强化**
|
|||
|
|
- 确保有清晰的标题层级（H1/H2/H3）
|
|||
|
|
- 在开头添加结论摘要（80-120字，概括核心观点）
|
|||
|
|
- 添加清单列表：将要点整理为清单格式（- 或 1. 格式）
|
|||
|
|
- 添加FAQ部分：至少3-5个常见问题解答
|
|||
|
|
- 添加代码块：技术内容添加代码示例（如适用，用占位符）
|
|||
|
|
- 添加对比表格：如有多个选项，用表格或列表对比
|
|||
|
|
- 添加步骤说明：如有流程，用步骤格式说明（1. 2. 3.）
|
|||
|
|
- 在结尾添加总结部分：总结核心观点和行动建议
|
|||
|
|
|
|||
|
|
【强化原则】
|
|||
|
|
- 保持原意和核心信息不变
|
|||
|
|
- 事实信息使用占位符，不要编造具体数据
|
|||
|
|
- 结构化块要自然融入，不要生硬插入
|
|||
|
|
- 长度控制在原长度的1.2-1.5倍
|
|||
|
|
|
|||
|
|
【输出格式】
|
|||
|
|
请输出两部分：
|
|||
|
|
|
|||
|
|
【强化后内容】
|
|||
|
|
（完整的优化后内容，增强事实密度和结构化块）
|
|||
|
|
|
|||
|
|
【强化说明】
|
|||
|
|
（列出添加的事实信息和结构化块，格式：类型 - 内容说明）
|
|||
|
|
|
|||
|
|
【开始优化】
|
|||
|
|
"""
|
|||
|
|
|
|||
|
|
def assess_fact_density(
|
|||
|
|
self,
|
|||
|
|
content: str,
|
|||
|
|
brand: str,
|
|||
|
|
advantages: str,
|
|||
|
|
platform: str,
|
|||
|
|
llm_chain
|
|||
|
|
) -> Dict:
|
|||
|
|
"""
|
|||
|
|
评估内容的事实密度和结构化程度
|
|||
|
|
|
|||
|
|
Args:
|
|||
|
|
content: 要评估的内容
|
|||
|
|
brand: 品牌名称
|
|||
|
|
advantages: 品牌优势
|
|||
|
|
platform: 发布平台
|
|||
|
|
llm_chain: LangChain 链对象
|
|||
|
|
|
|||
|
|
Returns:
|
|||
|
|
包含评估结果的字典
|
|||
|
|
"""
|
|||
|
|
try:
|
|||
|
|
prompt = PromptTemplate.from_template(self.assessment_prompt_template)
|
|||
|
|
chain = prompt | llm_chain | StrOutputParser()
|
|||
|
|
|
|||
|
|
result = chain.invoke({
|
|||
|
|
"content": content,
|
|||
|
|
"brand": brand,
|
|||
|
|
"advantages": advantages,
|
|||
|
|
"platform": platform
|
|||
|
|
})
|
|||
|
|
|
|||
|
|
# 解析结果
|
|||
|
|
assessment_data = self._parse_assessment_result(result)
|
|||
|
|
|
|||
|
|
return assessment_data
|
|||
|
|
|
|||
|
|
except Exception as e:
|
|||
|
|
# 如果评估失败，返回基于规则的评估
|
|||
|
|
return self._rule_based_assessment(content)
|
|||
|
|
|
|||
|
|
def enhance_fact_density(
|
|||
|
|
self,
|
|||
|
|
content: str,
|
|||
|
|
brand: str,
|
|||
|
|
advantages: str,
|
|||
|
|
platform: str,
|
|||
|
|
llm_chain
|
|||
|
|
) -> Dict:
|
|||
|
|
"""
|
|||
|
|
强化内容的事实密度和结构化程度
|
|||
|
|
|
|||
|
|
Args:
|
|||
|
|
content: 要强化的内容
|
|||
|
|
brand: 品牌名称
|
|||
|
|
advantages: 品牌优势
|
|||
|
|
platform: 发布平台
|
|||
|
|
llm_chain: LangChain 链对象
|
|||
|
|
|
|||
|
|
Returns:
|
|||
|
|
包含强化后内容和说明的字典
|
|||
|
|
"""
|
|||
|
|
try:
|
|||
|
|
prompt = PromptTemplate.from_template(self.enhancement_prompt_template)
|
|||
|
|
chain = prompt | llm_chain | StrOutputParser()
|
|||
|
|
|
|||
|
|
result = chain.invoke({
|
|||
|
|
"content": content,
|
|||
|
|
"brand": brand,
|
|||
|
|
"advantages": advantages,
|
|||
|
|
"platform": platform
|
|||
|
|
})
|
|||
|
|
|
|||
|
|
# 解析结果
|
|||
|
|
enhanced_data = self._parse_enhancement_result(result)
|
|||
|
|
|
|||
|
|
return enhanced_data
|
|||
|
|
|
|||
|
|
except Exception as e:
|
|||
|
|
return {
|
|||
|
|
"enhanced_content": content,
|
|||
|
|
"enhancement_details": [],
|
|||
|
|
"changes": f"事实密度强化失败：{str(e)}"
|
|||
|
|
}
|
|||
|
|
|
|||
|
|
def _parse_assessment_result(self, result: str) -> Dict:
|
|||
|
|
"""解析评估结果"""
|
|||
|
|
# 尝试提取 JSON
|
|||
|
|
json_match = re.search(r'\{.*\}', result, re.DOTALL)
|
|||
|
|
if json_match:
|
|||
|
|
try:
|
|||
|
|
data = json.loads(json_match.group())
|
|||
|
|
# 验证数据结构
|
|||
|
|
if "scores" in data and "total" in data["scores"]:
|
|||
|
|
return data
|
|||
|
|
except json.JSONDecodeError:
|
|||
|
|
pass
|
|||
|
|
|
|||
|
|
# 如果无法解析 JSON，使用基于规则的评估
|
|||
|
|
return self._rule_based_assessment("")
|
|||
|
|
|
|||
|
|
def _rule_based_assessment(self, content: str) -> Dict:
|
|||
|
|
"""
|
|||
|
|
基于规则的评估（备用方案）
|
|||
|
|
|
|||
|
|
Args:
|
|||
|
|
content: 内容文本
|
|||
|
|
|
|||
|
|
Returns:
|
|||
|
|
评估结果字典
|
|||
|
|
"""
|
|||
|
|
if not content:
|
|||
|
|
return {
|
|||
|
|
"scores": {
|
|||
|
|
"fact_density": 0,
|
|||
|
|
"structure": 0,
|
|||
|
|
"total": 0
|
|||
|
|
},
|
|||
|
|
"fact_analysis": {
|
|||
|
|
"data_count": 0,
|
|||
|
|
"case_count": 0,
|
|||
|
|
"standard_count": 0,
|
|||
|
|
"comparison_count": 0,
|
|||
|
|
"time_count": 0,
|
|||
|
|
"source_count": 0,
|
|||
|
|
"missing_facts": []
|
|||
|
|
},
|
|||
|
|
"structure_analysis": {
|
|||
|
|
"has_title": False,
|
|||
|
|
"has_summary": False,
|
|||
|
|
"has_list": False,
|
|||
|
|
"has_faq": False,
|
|||
|
|
"has_code": False,
|
|||
|
|
"has_table": False,
|
|||
|
|
"has_steps": False,
|
|||
|
|
"has_conclusion": False,
|
|||
|
|
"missing_blocks": []
|
|||
|
|
},
|
|||
|
|
"details": {
|
|||
|
|
"fact_density": "无法评估",
|
|||
|
|
"structure": "无法评估"
|
|||
|
|
},
|
|||
|
|
"improvements": ["请使用 LLM 进行详细评估"]
|
|||
|
|
}
|
|||
|
|
|
|||
|
|
# 事实密度检测
|
|||
|
|
data_patterns = [
|
|||
|
|
r'\d+%', r'\d+个', r'\d+项', r'约\d+', r'超过\d+', r'达到\d+',
|
|||
|
|
r'数据显示', r'统计', r'调研', r'报告'
|
|||
|
|
]
|
|||
|
|
case_patterns = [
|
|||
|
|
r'案例', r'实例', r'应用', r'实践', r'企业', r'公司'
|
|||
|
|
]
|
|||
|
|
standard_patterns = [
|
|||
|
|
r'标准', r'规范', r'要求', r'ISO', r'GB', r'行业'
|
|||
|
|
]
|
|||
|
|
comparison_patterns = [
|
|||
|
|
r'相比', r'对比', r'差异', r'优于', r'提升', r'降低'
|
|||
|
|
]
|
|||
|
|
time_patterns = [
|
|||
|
|
r'\d{4}年', r'最新', r'当前', r'目前', r'现在', r'今年'
|
|||
|
|
]
|
|||
|
|
source_patterns = [
|
|||
|
|
r'根据', r'参考', r'据', r'按照', r'显示', r'表明'
|
|||
|
|
]
|
|||
|
|
|
|||
|
|
data_count = sum(1 for p in data_patterns if re.search(p, content))
|
|||
|
|
case_count = sum(1 for p in case_patterns if re.search(p, content))
|
|||
|
|
standard_count = sum(1 for p in standard_patterns if re.search(p, content))
|
|||
|
|
comparison_count = sum(1 for p in comparison_patterns if re.search(p, content))
|
|||
|
|
time_count = sum(1 for p in time_patterns if re.search(p, content))
|
|||
|
|
source_count = sum(1 for p in source_patterns if re.search(p, content))
|
|||
|
|
|
|||
|
|
# 结构化检测
|
|||
|
|
has_title = bool(re.search(r'^#+\s+|^标题|^##', content, re.MULTILINE))
|
|||
|
|
has_summary = bool(re.search(r'摘要|总结|概述|概括', content[:200])) # 检查前200字
|
|||
|
|
has_list = bool(re.search(r'[-*•]\s+|^\d+[\.\)]\s+', content, re.MULTILINE))
|
|||
|
|
has_faq = bool(re.search(r'FAQ|常见问题|Q[：:]|问[：:]|答[：:]', content, re.IGNORECASE))
|
|||
|
|
has_code = bool(re.search(r'```|代码|示例|code', content, re.IGNORECASE))
|
|||
|
|
has_table = bool(re.search(r'\|.*\|', content) or re.search(r'表格|对比表', content))
|
|||
|
|
has_steps = bool(re.search(r'步骤|流程|第[一二三四五六七八九十\d]+[步点]', content))
|
|||
|
|
has_conclusion = bool(re.search(r'总结|结论|综上所述|总而言之', content[-200:])) # 检查后200字
|
|||
|
|
|
|||
|
|
# 计算分数
|
|||
|
|
fact_score = min(50, (data_count + case_count + standard_count +
|
|||
|
|
comparison_count + time_count + source_count) * 5)
|
|||
|
|
structure_score = sum([
|
|||
|
|
has_title * 7,
|
|||
|
|
has_summary * 7,
|
|||
|
|
has_list * 7,
|
|||
|
|
has_faq * 7,
|
|||
|
|
has_code * 6,
|
|||
|
|
has_table * 6,
|
|||
|
|
has_steps * 5,
|
|||
|
|
has_conclusion * 5
|
|||
|
|
])
|
|||
|
|
|
|||
|
|
# 缺失检测
|
|||
|
|
missing_facts = []
|
|||
|
|
if data_count == 0:
|
|||
|
|
missing_facts.append("数据信息")
|
|||
|
|
if case_count == 0:
|
|||
|
|
missing_facts.append("案例信息")
|
|||
|
|
if source_count == 0:
|
|||
|
|
missing_facts.append("来源信息")
|
|||
|
|
|
|||
|
|
missing_blocks = []
|
|||
|
|
if not has_title:
|
|||
|
|
missing_blocks.append("标题层级")
|
|||
|
|
if not has_summary:
|
|||
|
|
missing_blocks.append("结论摘要")
|
|||
|
|
if not has_list:
|
|||
|
|
missing_blocks.append("清单列表")
|
|||
|
|
if not has_faq:
|
|||
|
|
missing_blocks.append("FAQ部分")
|
|||
|
|
|
|||
|
|
return {
|
|||
|
|
"scores": {
|
|||
|
|
"fact_density": fact_score,
|
|||
|
|
"structure": structure_score,
|
|||
|
|
"total": fact_score + structure_score
|
|||
|
|
},
|
|||
|
|
"fact_analysis": {
|
|||
|
|
"data_count": data_count,
|
|||
|
|
"case_count": case_count,
|
|||
|
|
"standard_count": standard_count,
|
|||
|
|
"comparison_count": comparison_count,
|
|||
|
|
"time_count": time_count,
|
|||
|
|
"source_count": source_count,
|
|||
|
|
"missing_facts": missing_facts
|
|||
|
|
},
|
|||
|
|
"structure_analysis": {
|
|||
|
|
"has_title": has_title,
|
|||
|
|
"has_summary": has_summary,
|
|||
|
|
"has_list": has_list,
|
|||
|
|
"has_faq": has_faq,
|
|||
|
|
"has_code": has_code,
|
|||
|
|
"has_table": has_table,
|
|||
|
|
"has_steps": has_steps,
|
|||
|
|
"has_conclusion": has_conclusion,
|
|||
|
|
"missing_blocks": missing_blocks
|
|||
|
|
},
|
|||
|
|
"details": {
|
|||
|
|
"fact_density": f"检测到 {data_count + case_count + standard_count} 处事实信息",
|
|||
|
|
"structure": f"检测到 {sum([has_title, has_summary, has_list, has_faq, has_code, has_table, has_steps, has_conclusion])} 个结构化块"
|
|||
|
|
},
|
|||
|
|
"improvements": missing_facts + missing_blocks
|
|||
|
|
}
|
|||
|
|
|
|||
|
|
def _parse_enhancement_result(self, result: str) -> Dict:
|
|||
|
|
"""解析强化结果"""
|
|||
|
|
enhanced_content = ""
|
|||
|
|
enhancement_details = []
|
|||
|
|
|
|||
|
|
# 提取强化后内容
|
|||
|
|
if "【强化后内容】" in result:
|
|||
|
|
parts = result.split("【强化后内容】", 1)
|
|||
|
|
if len(parts) > 1:
|
|||
|
|
content_part = parts[1]
|
|||
|
|
if "【强化说明】" in content_part:
|
|||
|
|
enhanced_content = content_part.split("【强化说明】", 1)[0].strip()
|
|||
|
|
else:
|
|||
|
|
enhanced_content = content_part.strip()
|
|||
|
|
|
|||
|
|
# 提取强化说明
|
|||
|
|
if "【强化说明】" in result:
|
|||
|
|
detail_part = result.split("【强化说明】", 1)[1].strip()
|
|||
|
|
# 按行解析强化说明
|
|||
|
|
for line in detail_part.split("\n"):
|
|||
|
|
line = line.strip()
|
|||
|
|
if line and ("-" in line or "：" in line or ":" in line):
|
|||
|
|
enhancement_details.append(line)
|
|||
|
|
|
|||
|
|
# 如果没有找到明确的分隔符，尝试其他方式
|
|||
|
|
if not enhanced_content:
|
|||
|
|
enhanced_content = result.strip()
|
|||
|
|
|
|||
|
|
return {
|
|||
|
|
"enhanced_content": enhanced_content,
|
|||
|
|
"enhancement_details": enhancement_details,
|
|||
|
|
"changes": f"已添加 {len(enhancement_details)} 处事实信息和结构化块" if enhancement_details else "未检测到明确的强化内容"
|
|||
|
|
}
|
|||
|
|
|
|||
|
|
def get_score_level(self, total_score: int) -> tuple:
|
|||
|
|
"""
|
|||
|
|
根据总分返回等级和颜色
|
|||
|
|
|
|||
|
|
Returns:
|
|||
|
|
(等级名称, 颜色代码)
|
|||
|
|
"""
|
|||
|
|
if total_score >= 90:
|
|||
|
|
return ("优秀", "#10B981") # 绿色
|
|||
|
|
elif total_score >= 75:
|
|||
|
|
return ("良好", "#3B82F6") # 蓝色
|
|||
|
|
elif total_score >= 60:
|
|||
|
|
return ("中等", "#F59E0B") # 橙色
|
|||
|
|
else:
|
|||
|
|
return ("需改进", "#EF4444") # 红色
|