更新爬虫方案文档,增加摘要提取模块以生成文档摘要;优化基础爬虫类的标题提取逻辑,支持多个选择器,调整内容处理逻辑以去除重复标题。
This commit is contained in:
@@ -18,6 +18,7 @@ from abc import ABC, abstractmethod
|
||||
|
||||
from .config import BASE_URL, HEADERS, REQUEST_DELAY, OUTPUT_DIR
|
||||
from .utils import ensure_dir, download_image, safe_filename, make_absolute_url
|
||||
from .extract_abstract import generate_abstract
|
||||
|
||||
|
||||
class BaseCrawler(ABC):
|
||||
@@ -128,14 +129,28 @@ class BaseCrawler(ABC):
|
||||
selector = self.config.get("title_selector", "h1")
|
||||
index = self.config.get("title_index", 0)
|
||||
|
||||
# 支持多个选择器,用逗号分隔(类似 extract_content 的处理方式)
|
||||
selectors = [s.strip() for s in selector.split(',')]
|
||||
|
||||
# 收集所有匹配的标签
|
||||
all_tags = []
|
||||
for sel in selectors:
|
||||
# 对于简单的标签名(如 "h1", "h2"),直接查找
|
||||
if sel in ['h1', 'h2', 'h3', 'h4', 'h5', 'h6', 'title']:
|
||||
found_tags = soup.find_all(sel)
|
||||
all_tags.extend(found_tags)
|
||||
else:
|
||||
# 对于其他选择器,尝试查找
|
||||
found_tags = soup.find_all(sel)
|
||||
all_tags.extend(found_tags)
|
||||
|
||||
# 优先从配置的选择器提取
|
||||
tags = soup.find_all(selector)
|
||||
if tags and len(tags) > index:
|
||||
title = tags[index].get_text(strip=True)
|
||||
if all_tags and len(all_tags) > index:
|
||||
title = all_tags[index].get_text(strip=True)
|
||||
if title:
|
||||
return title
|
||||
elif tags:
|
||||
title = tags[0].get_text(strip=True)
|
||||
elif all_tags:
|
||||
title = all_tags[0].get_text(strip=True)
|
||||
if title:
|
||||
return title
|
||||
|
||||
@@ -328,19 +343,52 @@ class BaseCrawler(ABC):
|
||||
|
||||
return images_info
|
||||
|
||||
def content_to_markdown(self, content: BeautifulSoup) -> str:
|
||||
def content_to_markdown(self, content: BeautifulSoup, page_title: str = None) -> str:
|
||||
"""
|
||||
将内容转换为 Markdown
|
||||
|
||||
Args:
|
||||
content: 内容区域
|
||||
page_title: 页面标题(如果提供,会移除内容中与标题重复的第一个h1/h2标签)
|
||||
|
||||
Returns:
|
||||
Markdown 文本
|
||||
"""
|
||||
# 如果提供了页面标题,检查并移除内容中与标题重复的标签
|
||||
if page_title:
|
||||
# 创建内容的副本,避免修改原始内容
|
||||
content_copy = BeautifulSoup(str(content), 'html.parser')
|
||||
|
||||
# 移除与标题完全相同的第一个h1
|
||||
first_h1 = content_copy.find('h1')
|
||||
if first_h1:
|
||||
h1_text = first_h1.get_text(strip=True)
|
||||
if h1_text == page_title:
|
||||
first_h1.decompose()
|
||||
|
||||
# 移除与标题完全相同的第一个h2
|
||||
first_h2 = content_copy.find('h2')
|
||||
if first_h2:
|
||||
h2_text = first_h2.get_text(strip=True)
|
||||
if h2_text == page_title:
|
||||
first_h2.decompose()
|
||||
|
||||
# 检查标题是否包含"型号:"前缀,如果是,也移除内容中只包含产品名称的h2
|
||||
# 例如:标题是"型号:eCoder11",内容中有"eCoder11"的h2
|
||||
if '型号:' in page_title or '型号:' in page_title:
|
||||
product_name = page_title.replace('型号:', '').replace('型号:', '').strip()
|
||||
if product_name:
|
||||
# 查找第一个只包含产品名称的h2
|
||||
for h2 in content_copy.find_all('h2'):
|
||||
h2_text = h2.get_text(strip=True)
|
||||
if h2_text == product_name:
|
||||
h2.decompose()
|
||||
break # 只移除第一个匹配的
|
||||
|
||||
return markdownify.markdownify(str(content_copy), heading_style="ATX")
|
||||
return markdownify.markdownify(str(content), heading_style="ATX")
|
||||
|
||||
def add_content_to_docx(self, doc: Document, content: BeautifulSoup, output_dir: str):
|
||||
def add_content_to_docx(self, doc: Document, content: BeautifulSoup, output_dir: str, page_title: str = None):
|
||||
"""
|
||||
将内容添加到 Word 文档
|
||||
|
||||
@@ -348,7 +396,17 @@ class BaseCrawler(ABC):
|
||||
doc: Document 对象
|
||||
content: 内容区域
|
||||
output_dir: 输出目录(用于解析图片路径)
|
||||
page_title: 页面标题(如果提供,会跳过内容中与标题重复的第一个h1标签)
|
||||
"""
|
||||
# 如果提供了页面标题,创建内容副本并移除重复的h1
|
||||
if page_title:
|
||||
content = BeautifulSoup(str(content), 'html.parser')
|
||||
first_h1 = content.find('h1')
|
||||
if first_h1:
|
||||
h1_text = first_h1.get_text(strip=True)
|
||||
if h1_text == page_title:
|
||||
first_h1.decompose() # 移除该标签
|
||||
|
||||
# 按文档顺序处理元素,保持列表的连续性
|
||||
for element in content.find_all(['p', 'h1', 'h2', 'h3', 'h4', 'h5', 'h6', 'img', 'li', 'ul', 'ol', 'table']):
|
||||
if element.name == 'img':
|
||||
@@ -444,8 +502,8 @@ class BaseCrawler(ABC):
|
||||
# 处理图片
|
||||
images = self.process_images(content, url)
|
||||
|
||||
# 转换为 Markdown
|
||||
markdown = self.content_to_markdown(content)
|
||||
# 转换为 Markdown(传入标题,用于去除重复的h1标签)
|
||||
markdown = self.content_to_markdown(content, title)
|
||||
|
||||
return {
|
||||
"url": url,
|
||||
@@ -481,7 +539,7 @@ class BaseCrawler(ABC):
|
||||
p = doc.add_paragraph()
|
||||
p.add_run(f"原文链接: {page_data['url']}").italic = True
|
||||
|
||||
self.add_content_to_docx(doc, page_data["content"], self.output_dir)
|
||||
self.add_content_to_docx(doc, page_data["content"], self.output_dir, title)
|
||||
doc.save(docx_path)
|
||||
|
||||
def save_combined_documents(self, all_pages: list[dict]):
|
||||
@@ -505,14 +563,37 @@ class BaseCrawler(ABC):
|
||||
# === 处理 Markdown ===
|
||||
existing_urls = set()
|
||||
existing_content = ""
|
||||
existing_pages = [] # 存储已存在的页面信息(用于重新生成摘要)
|
||||
|
||||
# 如果文件已存在,读取现有内容并提取已存在的URL
|
||||
# 如果文件已存在,读取现有内容并提取已存在的URL和页面信息
|
||||
if os.path.exists(md_path):
|
||||
with open(md_path, "r", encoding="utf-8") as f:
|
||||
existing_content = f.read()
|
||||
# 提取已存在的URL(用于去重)
|
||||
url_pattern = r'\*\*原文链接\*\*: (https?://[^\s\n]+)'
|
||||
existing_urls = set(re.findall(url_pattern, existing_content))
|
||||
|
||||
# 提取已存在的页面信息(标题、URL和部分内容),用于重新生成摘要
|
||||
# 匹配格式:## 标题\n\n**原文链接**: URL\n\n内容...
|
||||
# 使用更复杂的正则来匹配每个页面的完整内容块
|
||||
page_blocks = re.split(r'\n\n---\n\n', existing_content)
|
||||
for block in page_blocks:
|
||||
# 匹配页面标题和URL
|
||||
title_match = re.search(r'^##\s+([^\n]+)', block, re.MULTILINE)
|
||||
url_match = re.search(r'\*\*原文链接\*\*:\s+(https?://[^\s\n]+)', block)
|
||||
if title_match and url_match:
|
||||
title = title_match.group(1).strip()
|
||||
url = url_match.group(1).strip()
|
||||
# 提取内容部分(跳过标题和URL行)
|
||||
content_start = url_match.end()
|
||||
markdown_content = block[content_start:].strip()
|
||||
# 只取前500字符作为预览
|
||||
markdown_preview = markdown_content[:500] if len(markdown_content) > 500 else markdown_content
|
||||
existing_pages.append({
|
||||
'title': title,
|
||||
'url': url,
|
||||
'markdown': markdown_preview
|
||||
})
|
||||
|
||||
# 过滤掉已存在的页面(基于URL去重)
|
||||
new_pages = [page for page in all_pages if page['url'] not in existing_urls]
|
||||
@@ -529,14 +610,63 @@ class BaseCrawler(ABC):
|
||||
new_md_content += page["markdown"]
|
||||
new_md_content += "\n\n---\n\n"
|
||||
|
||||
# 合并所有页面(已存在的 + 新添加的),用于生成摘要
|
||||
all_pages_for_abstract = existing_pages + all_pages
|
||||
|
||||
# 生成摘要(新建文档时生成,追加新内容时也重新生成,确保包含所有URL)
|
||||
abstract = None
|
||||
if not existing_content:
|
||||
# 新建文档:使用当前爬取的页面生成摘要
|
||||
print(f" 正在生成文档摘要...")
|
||||
abstract = generate_abstract(all_pages, output_dir_name)
|
||||
else:
|
||||
# 追加模式:重新生成摘要,包含所有页面(已存在的 + 新添加的)
|
||||
print(f" 正在重新生成文档摘要(包含所有 {len(all_pages_for_abstract)} 篇)...")
|
||||
abstract = generate_abstract(all_pages_for_abstract, output_dir_name)
|
||||
|
||||
# 追加或创建文件
|
||||
if existing_content:
|
||||
# 追加模式:在现有内容后追加新内容
|
||||
combined_md = existing_content.rstrip() + "\n\n" + new_md_content
|
||||
print(f" 追加 {len(new_pages)} 篇新内容到现有文档")
|
||||
# 追加模式:更新摘要部分,然后在现有内容后追加新内容
|
||||
# 使用正则表达式替换摘要部分(从标题后到第一个"---"分隔符之间的内容)
|
||||
# 匹配模式:标题行 + 摘要内容 + 分隔符
|
||||
title_pattern = r'^#\s+.*?全集\s*\n\n'
|
||||
separator_pattern = r'\n\n---\n\n'
|
||||
|
||||
# 查找标题后的第一个分隔符位置
|
||||
title_match = re.search(title_pattern, existing_content, re.MULTILINE)
|
||||
if title_match:
|
||||
title_end = title_match.end()
|
||||
# 查找第一个分隔符
|
||||
separator_match = re.search(separator_pattern, existing_content[title_end:])
|
||||
if separator_match:
|
||||
# 替换摘要部分
|
||||
separator_start = title_end + separator_match.start()
|
||||
separator_end = title_end + separator_match.end()
|
||||
# 保留标题和分隔符,替换中间的内容
|
||||
combined_md = existing_content[:title_end]
|
||||
if abstract:
|
||||
combined_md += abstract
|
||||
combined_md += existing_content[separator_end:]
|
||||
# 追加新内容
|
||||
combined_md = combined_md.rstrip() + "\n\n" + new_md_content
|
||||
else:
|
||||
# 如果没有找到分隔符,说明可能没有摘要,直接添加摘要和新内容
|
||||
combined_md = existing_content[:title_end]
|
||||
if abstract:
|
||||
combined_md += abstract + "\n\n---\n\n"
|
||||
combined_md += existing_content[title_end:].lstrip()
|
||||
combined_md = combined_md.rstrip() + "\n\n" + new_md_content
|
||||
else:
|
||||
# 如果没有找到标题,说明格式异常,直接追加
|
||||
combined_md = existing_content.rstrip() + "\n\n" + new_md_content
|
||||
print(f" 追加 {len(new_pages)} 篇新内容到现有文档,并更新摘要")
|
||||
else:
|
||||
# 新建模式:创建新文档
|
||||
combined_md = f"# {output_dir_name}全集\n\n" + new_md_content
|
||||
# 构建文档内容:标题 + 摘要 + 正文
|
||||
combined_md = f"# {output_dir_name}全集\n\n"
|
||||
if abstract:
|
||||
combined_md += f"{abstract}\n\n---\n\n"
|
||||
combined_md += new_md_content
|
||||
|
||||
with open(md_path, "w", encoding="utf-8") as f:
|
||||
f.write(combined_md)
|
||||
@@ -563,7 +693,7 @@ class BaseCrawler(ABC):
|
||||
doc.add_heading(page["title"], level=1)
|
||||
p = doc.add_paragraph()
|
||||
p.add_run(f"原文链接: {page['url']}").italic = True
|
||||
self.add_content_to_docx(doc, page["content"], self.output_dir)
|
||||
self.add_content_to_docx(doc, page["content"], self.output_dir, page["title"])
|
||||
doc.add_page_break()
|
||||
doc.save(docx_path)
|
||||
print(f" 追加 {len(new_pages_for_doc)} 篇新内容到 Word 文档")
|
||||
@@ -574,11 +704,28 @@ class BaseCrawler(ABC):
|
||||
doc = Document()
|
||||
doc.add_heading(f'{output_dir_name}全集', level=1)
|
||||
|
||||
# 添加摘要(只在新建时生成,复用Markdown部分生成的摘要)
|
||||
if not existing_content and abstract:
|
||||
# 将Markdown格式的摘要转换为Word格式
|
||||
# 处理Markdown链接:将 [文本](URL) 转换为 "文本 (URL)" 格式
|
||||
abstract_text = re.sub(r'\[([^\]]+)\]\(([^\)]+)\)', r'\1 (\2)', abstract)
|
||||
# 移除Markdown加粗标记
|
||||
abstract_text = abstract_text.replace('**', '')
|
||||
# 添加摘要段落
|
||||
for line in abstract_text.split('\n'):
|
||||
if line.strip():
|
||||
doc.add_paragraph(line.strip())
|
||||
else:
|
||||
doc.add_paragraph() # 空行
|
||||
doc.add_paragraph() # 空行
|
||||
doc.add_paragraph("─" * 50) # 分隔线
|
||||
doc.add_paragraph() # 空行
|
||||
|
||||
for page in all_pages:
|
||||
doc.add_heading(page["title"], level=1)
|
||||
p = doc.add_paragraph()
|
||||
p.add_run(f"原文链接: {page['url']}").italic = True
|
||||
self.add_content_to_docx(doc, page["content"], self.output_dir)
|
||||
self.add_content_to_docx(doc, page["content"], self.output_dir, page["title"])
|
||||
doc.add_page_break()
|
||||
|
||||
doc.save(docx_path)
|
||||
|
||||
91
zeroerr_crawler/extract_abstract.py
Normal file
91
zeroerr_crawler/extract_abstract.py
Normal file
@@ -0,0 +1,91 @@
|
||||
"""
|
||||
摘要提取模块
|
||||
使用大模型生成文档摘要
|
||||
"""
|
||||
|
||||
from openai import OpenAI
|
||||
|
||||
# API 配置
|
||||
API_BASE_URL = "https://yiming.zeroerr.team/v1"
|
||||
API_KEY = "sk-LX1g8KkG61S6eUaVD567C0C187D4452c90F9E6985cDf3586"
|
||||
MODEL = "Yiming"
|
||||
|
||||
|
||||
def generate_abstract(all_pages: list[dict], category_name: str) -> str:
|
||||
"""
|
||||
使用大模型生成文档摘要
|
||||
|
||||
Args:
|
||||
all_pages: 所有页面数据列表,每个元素包含 'title', 'url', 'markdown' 等字段
|
||||
category_name: 文档类别名称(如"应用案例")
|
||||
|
||||
Returns:
|
||||
摘要文本(Markdown格式),包含摘要内容和链接列表
|
||||
"""
|
||||
if not all_pages:
|
||||
return ""
|
||||
|
||||
try:
|
||||
# 构建文档内容(用于生成摘要)
|
||||
# 只使用标题和部分内容,避免内容过长
|
||||
content_parts = []
|
||||
for page in all_pages:
|
||||
title = page.get('title', '')
|
||||
markdown = page.get('markdown', '')
|
||||
# 只取前500字符的内容,避免输入过长
|
||||
content_preview = markdown[:500] if len(markdown) > 500 else markdown
|
||||
content_parts.append(f"标题:{title}\n内容预览:{content_preview}")
|
||||
|
||||
document_content = "\n\n".join(content_parts)
|
||||
|
||||
# 构建提示词
|
||||
prompt = f"""面向客户售前咨询,请为以下"{category_name}"类别的文档集合生成一个简洁的摘要。
|
||||
|
||||
文档内容:
|
||||
{document_content}
|
||||
|
||||
要求:
|
||||
1. 摘要应概括该页面的主题和主要内容
|
||||
2. 摘要长度控制在100-200字之间
|
||||
3. 使用简洁、专业的语言
|
||||
4. 突出该页面主题的价值和特点
|
||||
|
||||
请直接输出摘要内容,不要包含其他说明文字。"""
|
||||
|
||||
# 调用大模型API
|
||||
client = OpenAI(
|
||||
base_url=API_BASE_URL,
|
||||
api_key=API_KEY
|
||||
)
|
||||
|
||||
response = client.chat.completions.create(
|
||||
model=MODEL,
|
||||
temperature=0.3, # 使用较低的温度值,保证摘要的准确性
|
||||
messages=[
|
||||
{"role": "user", "content": prompt}
|
||||
]
|
||||
)
|
||||
|
||||
abstract_text = response.choices[0].message.content.strip()
|
||||
|
||||
# 构建链接列表
|
||||
links_section = "\n\n**相关链接:**\n\n"
|
||||
for i, page in enumerate(all_pages, 1):
|
||||
title = page.get('title', '未命名')
|
||||
url = page.get('url', '')
|
||||
links_section += f"{i}. [{title}]({url})\n"
|
||||
|
||||
# 组合摘要和链接
|
||||
result = f"{abstract_text}{links_section}"
|
||||
|
||||
return result
|
||||
|
||||
except Exception as e:
|
||||
print(f" 警告: 生成摘要失败: {e}")
|
||||
# 如果生成摘要失败,至少返回链接列表
|
||||
links_section = "\n\n**相关链接:**\n\n"
|
||||
for i, page in enumerate(all_pages, 1):
|
||||
title = page.get('title', '未命名')
|
||||
url = page.get('url', '')
|
||||
links_section += f"{i}. [{title}]({url})\n"
|
||||
return links_section
|
||||
@@ -45,24 +45,97 @@ class ProductCrawler(BaseCrawler):
|
||||
提取产品页面标题
|
||||
产品页面标题可能在不同位置
|
||||
"""
|
||||
# 尝试从面包屑导航后的第一个 h1
|
||||
h1_tags = soup.find_all('h1')
|
||||
for h1 in h1_tags:
|
||||
text = h1.get_text(strip=True)
|
||||
# 跳过网站名称
|
||||
if '零差云控' in text or '零误差' in text:
|
||||
continue
|
||||
if text:
|
||||
return text
|
||||
# 优先使用配置中的选择器(支持 h1, h2 等)
|
||||
selector = self.config.get("title_selector", "h1")
|
||||
index = self.config.get("title_index", 0)
|
||||
|
||||
# 从 URL 提取
|
||||
return url.split('/')[-1].replace('.html', '')
|
||||
# 支持多个选择器,用逗号分隔
|
||||
selectors = [s.strip() for s in selector.split(',')]
|
||||
|
||||
# 收集所有匹配的标签
|
||||
all_tags = []
|
||||
for sel in selectors:
|
||||
# 对于简单的标签名(如 "h1", "h2"),直接查找
|
||||
if sel in ['h1', 'h2', 'h3', 'h4', 'h5', 'h6', 'title']:
|
||||
found_tags = soup.find_all(sel)
|
||||
all_tags.extend(found_tags)
|
||||
else:
|
||||
# 对于其他选择器,尝试查找
|
||||
found_tags = soup.find_all(sel)
|
||||
all_tags.extend(found_tags)
|
||||
|
||||
# 优先从配置的选择器提取
|
||||
if all_tags and len(all_tags) > index:
|
||||
title = all_tags[index].get_text(strip=True)
|
||||
# 跳过网站名称
|
||||
if title and '零差云控' not in title and '零误差' not in title:
|
||||
return title
|
||||
elif all_tags:
|
||||
# 如果指定索引的标签被跳过,尝试其他标签
|
||||
for tag in all_tags:
|
||||
title = tag.get_text(strip=True)
|
||||
# 跳过网站名称
|
||||
if title and '零差云控' not in title and '零误差' not in title:
|
||||
return title
|
||||
|
||||
# 尝试从页面 title 标签提取
|
||||
title_tag = soup.find('title')
|
||||
if title_tag:
|
||||
title = title_tag.get_text(strip=True)
|
||||
# 移除网站名称后缀(如 " - 零差云控")
|
||||
if ' - ' in title:
|
||||
title = title.split(' - ')[0].strip()
|
||||
if title and title.lower() not in ['about-us', 'contact-us', 'join-us']:
|
||||
return title
|
||||
|
||||
# 最后从 URL 提取
|
||||
url_part = url.split('/')[-1].replace('.html', '')
|
||||
# 将连字符替换为空格,并首字母大写
|
||||
if '-' in url_part:
|
||||
url_part = ' '.join(word.capitalize() for word in url_part.split('-'))
|
||||
return url_part
|
||||
|
||||
def add_content_to_docx(self, doc: Document, content: BeautifulSoup, output_dir: str):
|
||||
def add_content_to_docx(self, doc: Document, content: BeautifulSoup, output_dir: str, page_title: str = None):
|
||||
"""
|
||||
将产品内容添加到 Word 文档
|
||||
针对产品页面的表格等进行优化处理
|
||||
|
||||
Args:
|
||||
doc: Document 对象
|
||||
content: 内容区域
|
||||
output_dir: 输出目录(用于解析图片路径)
|
||||
page_title: 页面标题(如果提供,会跳过内容中与标题重复的h1/h2标签或包含标题的段落)
|
||||
"""
|
||||
# 如果提供了页面标题,创建内容副本并移除重复的标题元素
|
||||
if page_title:
|
||||
content = BeautifulSoup(str(content), 'html.parser')
|
||||
|
||||
# 移除与标题完全相同的第一个h1
|
||||
first_h1 = content.find('h1')
|
||||
if first_h1:
|
||||
h1_text = first_h1.get_text(strip=True)
|
||||
if h1_text == page_title:
|
||||
first_h1.decompose()
|
||||
|
||||
# 移除与标题完全相同的第一个h2
|
||||
first_h2 = content.find('h2')
|
||||
if first_h2:
|
||||
h2_text = first_h2.get_text(strip=True)
|
||||
if h2_text == page_title:
|
||||
first_h2.decompose()
|
||||
|
||||
# 检查标题是否包含"型号:"前缀,如果是,也移除内容中只包含产品名称的h2
|
||||
# 例如:标题是"型号:eCoder11",内容中有"eCoder11"的h2
|
||||
if '型号:' in page_title or '型号:' in page_title:
|
||||
product_name = page_title.replace('型号:', '').replace('型号:', '').strip()
|
||||
if product_name:
|
||||
# 查找第一个只包含产品名称的h2
|
||||
for h2 in content.find_all('h2'):
|
||||
h2_text = h2.get_text(strip=True)
|
||||
if h2_text == product_name:
|
||||
h2.decompose()
|
||||
break # 只移除第一个匹配的
|
||||
|
||||
for element in content.find_all(['p', 'h1', 'h2', 'h3', 'h4', 'h5', 'h6', 'img', 'li', 'table', 'div']):
|
||||
# 跳过嵌套元素
|
||||
if element.find_parent(['table', 'li']):
|
||||
|
||||
Reference in New Issue
Block a user