文章目录
- 发现宝藏
- 一、 目标
- 二、 浅析
- 三、获取所有模块
- 四、请求处理模块、版面、文章
- 分析切换页面的参数传递- 2. 获取共有多少页标签并遍历版面- 3.解析版面并保存版面信息- 4. 解析文章列表和文章- 5. 清洗文章- 6. 保存文章图片
- 五、完整代码
- 六、效果展示
发现宝藏
前些天发现了一个巨牛的人工智能学习网站,通俗易懂,风趣幽默,忍不住分享一下给大家。【宝藏入口】。
一、 目标
爬取
news.mit.edu
的字段,包含标题、内容,作者,发布时间,链接地址,文章快照 (可能需要翻墙才能访问)
二、 浅析
1.全部新闻大致分为4个模块
2.每个模块的标签列表大致如下
3.每个标签对应的文章列表大致如下
4.具体每篇文章对应的结构如下
三、获取所有模块
其实就四个模块,列举出来就好,然后对每个分别解析爬取每个模块
classMitnewsScraper:def__init__(self, root_url, model_url, img_output_dir):
self.root_url = root_url
self.model_url = model_url
self.img_output_dir = img_output_dir
self.headers ={'Referer':'https://news.mit.edu/','User-Agent':'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) ''Chrome/122.0.0.0 Safari/537.36','Cookie':'替换成你自己的',}...defrun():
root_url ='https://news.mit.edu/'
model_urls =['https://news.mit.edu/topic','https://news.mit.edu/clp','https://news.mit.edu/department','https://news.mit.edu/']
output_dir ='D:\imgs\mit-news'for model_url in model_urls:
scraper = MitnewsScraper(root_url, model_url, output_dir)
scraper.catalogue_all_pages()
四、请求处理模块、版面、文章
先处理一个模块(TOPICS)
1. 分析切换页面的参数传递
如图可知是get请求,需要传一个参数page
2. 获取共有多少页标签并遍历版面
实际上是获取所有的page参数,然后进行遍历获取所有的标签
# 获取一个模块有多少版面defcatalogue_all_pages(self):
response = requests.get(self.model_url, headers=self.headers)
soup = BeautifulSoup(response.text,'html.parser')try:match= re.search(r'of (\d+) topics', soup.text)
total_catalogues =int(match.group(1))
total_pages = math.ceil(total_catalogues /20)print('topics模块一共有'+match.group(1)+'个版面,'+str(total_pages)+'页数据')for page inrange(0, total_pages):
self.parse_catalogues(page)print(f"========Finished catalogues page {page +1}========")except:
self.parse_catalogues(0)
3.解析版面并保存版面信息
前三个模块的版面列表
第四个模块的版面列表
# 解析版面列表里的版面defparse_catalogues(self, page):
params ={'page': page}
response = requests.get(self.model_url, params=params, headers=self.headers)if response.status_code ==200:
soup = BeautifulSoup(response.text,'html.parser')if self.root_url == self.model_url:
catalogue_list = soup.find('div','site-browse--recommended-section site-browse--recommended-section--schools')
catalogues_list = catalogue_list.find_all('li')else:
catalogue_list = soup.find('ul','page-vocabulary--views--list')
catalogues_list = catalogue_list.find_all('li')for index, catalogue inenumerate(catalogues_list):# 操作时间
date = datetime.now()# 版面标题
catalogue_title = catalogue.find('a').get_text(strip=True)print('第'+str(index +1)+'个版面标题为:'+ catalogue_title)
catalogue_href = catalogue.find('a').get('href')# 版面id
catalogue_id = catalogue_href[1:]
catalogue_url = self.root_url + catalogue_href
print('第'+str(index +1)+'个版面地址为:'+ catalogue_url)# 根据版面url解析文章列表
response = requests.get(catalogue_url, headers=self.headers)
soup = BeautifulSoup(response.text,'html.parser')match= re.search(r'of (\d+)', soup.text)# 查找一个版面有多少篇文章
total_cards =int(match.group(1))
total_pages = math.ceil(total_cards /15)print(f'{catalogue_title}版面一共有{total_cards}篇文章,'+f'{total_pages}页数据')for page inrange(0, total_pages):
self.parse_cards_list(page, catalogue_url, catalogue_id)print(f"========Finished {catalogue_title} 版面 page {page +1}========")# 连接 MongoDB 数据库服务器
client = MongoClient('mongodb://localhost:27017/')# 创建或选择数据库
db = client['mit-news']# 创建或选择集合
catalogues_collection = db['catalogues']# 插入示例数据到 catalogues 集合
catalogue_data ={'id': catalogue_id,'date': date,'title': catalogue_title,'url': catalogue_url,'cardSize': total_cards
}
catalogues_collection.insert_one(catalogue_data)returnTrueelse:raise Exception(f"Failed to fetch page {page}. Status code: {response.status_code}")
4. 解析文章列表和文章
寻找冗余部分并删除,例如
# 解析文章列表里的文章defparse_cards_list(self, page, url, catalogue_id):
params ={'page': page}
response = requests.get(url, params=params, headers=self.headers)if response.status_code ==200:
soup = BeautifulSoup(response.text,'html.parser')
card_list = soup.find('div','page-term--views--list')
cards_list = card_list.find_all('div','page-term--views--list-item')for index, card inenumerate(cards_list):# 对应的版面id
catalogue_id = catalogue_id
# 操作时间
date = datetime.now()# 文章标题
card_title = card.find('a','term-page--news-article--item--title--link').find('span').get_text(
strip=True)# 文章简介
card_introduction = card.find('p','term-page--news-article--item--dek').find('span').get_text(
strip=True)# 文章更新时间
publish_time = card.find('p','term-page--news-article--item--publication-date').find('time').get('datetime')
updateTime = datetime.strptime(publish_time,'%Y-%m-%dT%H:%M:%SZ')# 文章地址
temp_url = card.find('div','term-page--news-article--item--cover-image').find('a').get('href')
url ='https://news.mit.edu'+ temp_url
# 文章id
pattern =r'(\w+(-\w+)*)-(\d+)'match= re.search(pattern, temp_url)
card_id =str(match.group(0))
card_response = requests.get(url, headers=self.headers)
soup = BeautifulSoup(card_response.text,'html.parser')# 原始htmldom结构
html_title = soup.find('div',id='block-mit-page-title')
html_content = soup.find('div',id='block-mit-content')# 合并标题和内容
html_title.append(html_content)
html_cut1 = soup.find('div','news-article--topics')
html_cut2 = soup.find('div','news-article--archives')
html_cut3 = soup.find('div','news-article--content--side-column')
html_cut4 = soup.find('div','news-article--press-inquiries')
html_cut5 = soup.find_all('div','visually-hidden')
html_cut6 = soup.find('p','news-article--images-gallery--nav--inner')# 移除元素if html_cut1:
html_cut1.extract()if html_cut2:
html_cut2.extract()if html_cut3:
html_cut3.extract()if html_cut4:
html_cut4.extract()if html_cut5:for item in html_cut5:
item.extract()if html_cut6:
html_cut6.extract()# 获取合并后的内容文本
html_content = html_title
# 文章作者
author_list = html_content.find('div','news-article--authored-by').find_all('span')
author =''for item in author_list:
author = author + item.get_text()# 增加保留html样式的源文本
origin_html = html_content.prettify()# String# 转义网页中的图片标签
str_html = self.transcoding_tags(origin_html)# 再包装成
temp_soup = BeautifulSoup(str_html,'html.parser')# 反转译文件中的插图
str_html = self.translate_tags(temp_soup.text)# 绑定更新内容
content = self.clean_content(str_html)# 下载图片
imgs =[]
img_array = soup.find_all('div','news-article--image-item')for item in img_array:
img_url = self.root_url + item.find('img').get('data-src')
imgs.append(img_url)iflen(imgs)!=0:# 下载图片
illustrations = self.download_images(imgs, card_id)# 连接 MongoDB 数据库服务器
client = MongoClient('mongodb://localhost:27017/')# 创建或选择数据库
db = client['mit-news']# 创建或选择集合
cards_collection = db['cards']# 插入示例数据到 catalogues 集合
card_data ={'id': card_id,'catalogueId': catalogue_id,'type':'mit-news','date': date,'title': card_title,'author': author,'card_introduction': card_introduction,'updatetime': updateTime,'url': url,'html_content':str(html_content),'content': content,'illustrations': illustrations,}
cards_collection.insert_one(card_data)returnTrueelse:raise Exception(f"Failed to fetch page {page}. Status code: {response.status_code}")
5. 清洗文章
# 工具 转义标签deftranscoding_tags(self, htmlstr):
re_img = re.compile(r'\s*<(img.*?)>\s*', re.M)
s = re_img.sub(r'\n @@##\1##@@ \n', htmlstr)# IMG 转义return s
# 工具 转义标签deftranslate_tags(self, htmlstr):
re_img = re.compile(r'@@##(img.*?)##@@', re.M)
s = re_img.sub(r'<\1>', htmlstr)# IMG 转义return s
# 清洗文章defclean_content(self, content):if content isnotNone:
content = re.sub(r'\r',r'\n', content)
content = re.sub(r'\n{2,}','', content)
content = re.sub(r' {6,}','', content)
content = re.sub(r' {3,}\n','', content)
content = re.sub(r'<img src="../../../image/zxbl.gif"/>','', content)
content = content.replace('<img border="0" src="****处理标记:[Article]时, 字段 [SnapUrl] 在数据源中没有找到! ****"/> ','')
content = content.replace(''' <!--/enpcontent<INPUT type=checkbox value=0 name=titlecheckbox sourceid="<Source>SourcePh " style="display:none">''','') \
.replace(' <!--enpcontent','').replace('<TABLE>','')
content = content.replace('<P>','').replace('<\P>','').replace(' ',' ')return content
6. 保存文章图片
# 下载图片defdownload_images(self, img_urls, card_id):# 根据card_id创建一个新的子目录
images_dir = os.path.join(self.img_output_dir, card_id)ifnot os.path.exists(images_dir):
os.makedirs(images_dir)
downloaded_images =[]for index, img_url inenumerate(img_urls):try:
response = requests.get(img_url, stream=True, headers=self.headers)if response.status_code ==200:# 从URL中提取图片文件名
img_name_with_extension = img_url.split('/')[-1]
pattern =r'^[^?]*'match= re.search(pattern, img_name_with_extension)
img_name =match.group(0)# 保存图片withopen(os.path.join(images_dir, img_name),'wb')as f:
f.write(response.content)
downloaded_images.append([img_url, os.path.join(images_dir, img_name)])except requests.exceptions.RequestException as e:print(f'请求图片时发生错误:{e}')except Exception as e:print(f'保存图片时发生错误:{e}')return downloaded_images
# 如果文件夹存在则跳过else:print(f'文章id为{card_id}的图片文件夹已经存在')return[]
五、完整代码
import os
from datetime import datetime
import requests
from bs4 import BeautifulSoup
from pymongo import MongoClient
import re
import math
classMitnewsScraper:def__init__(self, root_url, model_url, img_output_dir):
self.root_url = root_url
self.model_url = model_url
self.img_output_dir = img_output_dir
self.headers ={'Referer':'https://news.mit.edu/','User-Agent':'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) ''Chrome/122.0.0.0 Safari/537.36','Cookie':'替换成你自己的'}# 获取一个模块有多少版面defcatalogue_all_pages(self):
response = requests.get(self.model_url, headers=self.headers)
soup = BeautifulSoup(response.text,'html.parser')try:match= re.search(r'of (\d+) topics', soup.text)
total_catalogues =int(match.group(1))
total_pages = math.ceil(total_catalogues /20)print('topics模块一共有'+match.group(1)+'个版面,'+str(total_pages)+'页数据')for page inrange(0, total_pages):
self.parse_catalogues(page)print(f"========Finished catalogues page {page +1}========")except:
self.parse_catalogues(0)# 解析版面列表里的版面defparse_catalogues(self, page):
params ={'page': page}
response = requests.get(self.model_url, params=params, headers=self.headers)if response.status_code ==200:
soup = BeautifulSoup(response.text,'html.parser')if self.root_url == self.model_url:
catalogue_list = soup.find('div','site-browse--recommended-section site-browse--recommended-section--schools')
catalogues_list = catalogue_list.find_all('li')else:
catalogue_list = soup.find('ul','page-vocabulary--views--list')
catalogues_list = catalogue_list.find_all('li')for index, catalogue inenumerate(catalogues_list):# 操作时间
date = datetime.now()# 版面标题
catalogue_title = catalogue.find('a').get_text(strip=True)print('第'+str(index +1)+'个版面标题为:'+ catalogue_title)
catalogue_href = catalogue.find('a').get('href')# 版面id
catalogue_id = catalogue_href[1:]
catalogue_url = self.root_url + catalogue_href
print('第'+str(index +1)+'个版面地址为:'+ catalogue_url)# 根据版面url解析文章列表
response = requests.get(catalogue_url, headers=self.headers)
soup = BeautifulSoup(response.text,'html.parser')match= re.search(r'of (\d+)', soup.text)# 查找一个版面有多少篇文章
total_cards =int(match.group(1))
total_pages = math.ceil(total_cards /15)print(f'{catalogue_title}版面一共有{total_cards}篇文章,'+f'{total_pages}页数据')for page inrange(0, total_pages):
self.parse_cards_list(page, catalogue_url, catalogue_id)print(f"========Finished {catalogue_title} 版面 page {page +1}========")# 连接 MongoDB 数据库服务器
client = MongoClient('mongodb://localhost:27017/')# 创建或选择数据库
db = client['mit-news']# 创建或选择集合
catalogues_collection = db['catalogues']# 插入示例数据到 catalogues 集合
catalogue_data ={'id': catalogue_id,'date': date,'title': catalogue_title,'url': catalogue_url,'cardSize': total_cards
}
catalogues_collection.insert_one(catalogue_data)returnTrueelse:raise Exception(f"Failed to fetch page {page}. Status code: {response.status_code}")# 解析文章列表里的文章defparse_cards_list(self, page, url, catalogue_id):
params ={'page': page}
response = requests.get(url, params=params, headers=self.headers)if response.status_code ==200:
soup = BeautifulSoup(response.text,'html.parser')
card_list = soup.find('div','page-term--views--list')
cards_list = card_list.find_all('div','page-term--views--list-item')for index, card inenumerate(cards_list):# 对应的版面id
catalogue_id = catalogue_id
# 操作时间
date = datetime.now()# 文章标题
card_title = card.find('a','term-page--news-article--item--title--link').find('span').get_text(
strip=True)# 文章简介
card_introduction = card.find('p','term-page--news-article--item--dek').find('span').get_text(
strip=True)# 文章更新时间
publish_time = card.find('p','term-page--news-article--item--publication-date').find('time').get('datetime')
updateTime = datetime.strptime(publish_time,'%Y-%m-%dT%H:%M:%SZ')# 文章地址
temp_url = card.find('div','term-page--news-article--item--cover-image').find('a').get('href')
url ='https://news.mit.edu'+ temp_url
# 文章id
pattern =r'(\w+(-\w+)*)-(\d+)'match= re.search(pattern, temp_url)
card_id =str(match.group(0))
card_response = requests.get(url, headers=self.headers)
soup = BeautifulSoup(card_response.text,'html.parser')# 原始htmldom结构
html_title = soup.find('div',id='block-mit-page-title')
html_content = soup.find('div',id='block-mit-content')# 合并标题和内容
html_title.append(html_content)
html_cut1 = soup.find('div','news-article--topics')
html_cut2 = soup.find('div','news-article--archives')
html_cut3 = soup.find('div','news-article--content--side-column')
html_cut4 = soup.find('div','news-article--press-inquiries')
html_cut5 = soup.find_all('div','visually-hidden')
html_cut6 = soup.find('p','news-article--images-gallery--nav--inner')# 移除元素if html_cut1:
html_cut1.extract()if html_cut2:
html_cut2.extract()if html_cut3:
html_cut3.extract()if html_cut4:
html_cut4.extract()if html_cut5:for item in html_cut5:
item.extract()if html_cut6:
html_cut6.extract()# 获取合并后的内容文本
html_content = html_title
# 文章作者
author_list = html_content.find('div','news-article--authored-by').find_all('span')
author =''for item in author_list:
author = author + item.get_text()# 增加保留html样式的源文本
origin_html = html_content.prettify()# String# 转义网页中的图片标签
str_html = self.transcoding_tags(origin_html)# 再包装成
temp_soup = BeautifulSoup(str_html,'html.parser')# 反转译文件中的插图
str_html = self.translate_tags(temp_soup.text)# 绑定更新内容
content = self.clean_content(str_html)# 下载图片
imgs =[]
img_array = soup.find_all('div','news-article--image-item')for item in img_array:
img_url = self.root_url + item.find('img').get('data-src')
imgs.append(img_url)iflen(imgs)!=0:# 下载图片
illustrations = self.download_images(imgs, card_id)# 连接 MongoDB 数据库服务器
client = MongoClient('mongodb://localhost:27017/')# 创建或选择数据库
db = client['mit-news']# 创建或选择集合
cards_collection = db['cards']# 插入示例数据到 catalogues 集合
card_data ={'id': card_id,'catalogueId': catalogue_id,'type':'mit-news','date': date,'title': card_title,'author': author,'card_introduction': card_introduction,'updatetime': updateTime,'url': url,'html_content':str(html_content),'content': content,'illustrations': illustrations,}
cards_collection.insert_one(card_data)returnTrueelse:raise Exception(f"Failed to fetch page {page}. Status code: {response.status_code}")# 下载图片defdownload_images(self, img_urls, card_id):# 根据card_id创建一个新的子目录
images_dir = os.path.join(self.img_output_dir, card_id)ifnot os.path.exists(images_dir):
os.makedirs(images_dir)
downloaded_images =[]for index, img_url inenumerate(img_urls):try:
response = requests.get(img_url, stream=True, headers=self.headers)if response.status_code ==200:# 从URL中提取图片文件名
img_name_with_extension = img_url.split('/')[-1]
pattern =r'^[^?]*'match= re.search(pattern, img_name_with_extension)
img_name =match.group(0)# 保存图片withopen(os.path.join(images_dir, img_name),'wb')as f:
f.write(response.content)
downloaded_images.append([img_url, os.path.join(images_dir, img_name)])except requests.exceptions.RequestException as e:print(f'请求图片时发生错误:{e}')except Exception as e:print(f'保存图片时发生错误:{e}')return downloaded_images
# 如果文件夹存在则跳过else:print(f'文章id为{card_id}的图片文件夹已经存在')return[]# 工具 转义标签deftranscoding_tags(self, htmlstr):
re_img = re.compile(r'\s*<(img.*?)>\s*', re.M)
s = re_img.sub(r'\n @@##\1##@@ \n', htmlstr)# IMG 转义return s
# 工具 转义标签deftranslate_tags(self, htmlstr):
re_img = re.compile(r'@@##(img.*?)##@@', re.M)
s = re_img.sub(r'<\1>', htmlstr)# IMG 转义return s
# 清洗文章defclean_content(self, content):if content isnotNone:
content = re.sub(r'\r',r'\n', content)
content = re.sub(r'\n{2,}','', content)
content = re.sub(r' {6,}','', content)
content = re.sub(r' {3,}\n','', content)
content = re.sub(r'<img src="../../../image/zxbl.gif"/>','', content)
content = content.replace('<img border="0" src="****处理标记:[Article]时, 字段 [SnapUrl] 在数据源中没有找到! ****"/> ','')
content = content.replace(''' <!--/enpcontent<INPUT type=checkbox value=0 name=titlecheckbox sourceid="<Source>SourcePh " style="display:none">''','') \
.replace(' <!--enpcontent','').replace('<TABLE>','')
content = content.replace('<P>','').replace('<\P>','').replace(' ',' ')return content
defrun():
root_url ='https://news.mit.edu/'
model_urls =['https://news.mit.edu/topic','https://news.mit.edu/clp','https://news.mit.edu/department','https://news.mit.edu/']
output_dir ='D:\imgs\mit-news'for model_url in model_urls:
scraper = MitnewsScraper(root_url, model_url, output_dir)
scraper.catalogue_all_pages()if __name__ =="__main__":
run()
六、效果展示
版权归原作者 东离与糖宝 所有, 如有侵权,请联系我们删除。