欢迎您访问程序员文章站本站旨在为大家提供分享程序员计算机编程知识!
您现在的位置是: 首页  >  IT编程

使用python实现抓取腾讯视频所有电影的爬虫

程序员文章站 2022-05-09 23:16:58
用python实现的抓取腾讯视频所有电影的爬虫 # -*- coding: utf-8 -*- import re import urllib2 from...

用python实现的抓取腾讯视频所有电影的爬虫

# -*- coding: utf-8 -*-
import re
import urllib2
from bs4import beautifulsoup
import string, time
import pymongo
num   =0 #全局变量,电影数量
m_type = u'' #全局变量,电影类型
m_site = u'qq' #全局变量,电影网站
#根据指定的url获取网页内容
def gethtml(url):
  req = urllib2.request(url)
  response = urllib2.urlopen(req)
  html = response.read()
  return html
#从电影分类列表页面获取电影分类
def gettags(html):
  global m_type
  soup = beautifulsoup(html)   #过滤出分类内容
  #print soup
  #<ulclass="clearfix _group" gname="mi_type" gtype="1">
  tags_all = soup.find_all('ul', {'class' :'clearfix _group' ,'gname' :'mi_type'})
  #print len(tags_all), tags_all
  #print str(tags_all[1]).replace('\n','')
  #<a _hot="tag.sub" class="_gtag _hotkey" href="http://v.qq.com/list/1_0_-1_-1_1_0_0_20_0_-1_0.html" title="动作" tvalue="0">动作</a>
  re_tags = r'<a _hot=\"tag\.sub\" class=\"_gtag _hotkey\" href=\"(.+?)\" title=\"(.+?)\" tvalue=\"(.+?)\">.+?</a>'
  p = re.compile(re_tags, re.dotall)
  tags = p.findall(str(tags_all[0]))
  if tags:
    tags_url = {}
    #print tags
    for tagin tags:
      tag_url = tag[0].decode('utf-8')
      #print tag_url
      m_type = tag[1].decode('utf-8')
      tags_url[m_type] = tag_url
  else:
      print"not find"
  return tags_url
#获取每个分类的页数
def get_pages(tag_url):
  tag_html = gethtml(tag_url)
  #divclass="paginator
  soup = beautifulsoup(tag_html)   #过滤出标记页面的html
  #print soup
  #<divclass="mod_pagenav" id="pager">
  div_page = soup.find_all('div', {'class' :'mod_pagenav','id' :'pager'})
  #print div_page #len(div_page), div_page[0]
  #<aclass="c_txt6" href="http://v.qq.com/list/1_2_-1_-1_1_0_24_20_0_-1_0.html" title="25"><span>25</span></a>
  re_pages = r'<a class=.+?><span>(.+?)</span></a>'
  p = re.compile(re_pages, re.dotall)
  pages = p.findall(str(div_page[0]))
  #print pages
  if len(pages) >1:
    return pages[-2]
  else:
    return 1
def getmovielist(html):
  soup = beautifulsoup(html)
  #<ulclass="mod_list_pic_130">
  divs = soup.find_all('ul', {'class' :'mod_list_pic_130'})
  #print divs
  for div_htmlin divs:
    div_html = str(div_html).replace('\n','')
    #print div_html
    getmovie(div_html)
def getmovie(html):
  global num
  global m_type
  global m_site
  re_movie = r'<li><a class=\"mod_poster_130\" href=\"(.+?)\" target=\"_blank\" title=\"(.+?)\"><img.+?</li>'
  p = re.compile(re_movie, re.dotall)
  movies = p.findall(html)
  if movies:
    conn = pymongo.connection('localhost',27017)
    movie_db = conn.dianying
    playlinks = movie_db.playlinks
    #print movies
    for moviein movies:
      #print movie
      num +=1
      print"%s : %d" % ("=" *70, num)
      values = dict(
        movie_title = movie[1],
        movie_url  = movie[0],
        movie_site   = m_site,
        movie_type   = m_type
        )
      print values
      playlinks.insert(values)
      print"_" *70
      num +=1
      print"%s : %d" % ("=" *70, num)
  #else:
  #  print"not find"
def getmovieinfo(url):
  html = gethtml(url)
  soup = beautifulsoup(html)
  #pack pack_album album_cover
  divs = soup.find_all('div', {'class' :'pack pack_album album_cover'})
  #print divs[0]
  #<a href="http://www.tudou.com/albumplay/9nyofxc_lhi/32jqhikjyki.html" target="new" title="《血滴子》独家纪录片" wl="1"> </a>
  re_info = r'<a href=\"(.+?)\" target=\"new\" title=\"(.+?)\" wl=\".+?\"> </a>'
  p_info = re.compile(re_info, re.dotall)
  m_info = p_info.findall(str(divs[0]))
  if m_info:
    return m_info
  else:
    print"not find movie info"
  return m_info
def insertdb(movieinfo):
  global conn
  movie_db = conn.dianying_at
  movies = movie_db.movies
  movies.insert(movieinfo)
if __name__ =="__main__":
  global conn
  tags_url ="http://v.qq.com/list/1_-1_-1_-1_1_0_0_20_0_-1_0.html"
  #print tags_url
  tags_html = gethtml(tags_url)
  #print tags_html
  tag_urls = gettags(tags_html)
  #print tag_urls
  for urlin tag_urls.items():
    print str(url[1]).encode('utf-8') #,url[0]
    maxpage =int(get_pages(str(url[1]).encode('utf-8')))
    print maxpage
    for xin range(0, maxpage):
      #http://v.qq.com/list/1_0_-1_-1_1_0_0_20_0_-1_0.html
      m_url = str(url[1]).replace('0_20_0_-1_0.html','')
      movie_url ="%s%d_20_0_-1_0.html" % (m_url, x)
      print movie_url
      movie_html = gethtml(movie_url.encode('utf-8'))
      #print movie_html
      getmovielist(movie_html)
      time.sleep(0.1)

总结

以上所述是小编给大家介绍的使用python实现抓取腾讯视频所有电影的爬虫,希望对大家有所帮助