本帖最后由 zhukeming123 于 2019-11-17 01:13 编辑
链家房源爬虫,基于scrapy框架,数据存储在MongoDB
由于链家的限制房源列表最多显示1000页,3000条数据,所以这里通过小区信息爬取所有房源。
# -*- coding: utf-8 -*-
import scrapy
from scrapy import Selector
import json
import re
from house_spider.items import LianjiaVillageItem, LianjiaHouseItem
class LianjiaSpider(scrapy.Spider):
name = 'lianjia'
allowed_domains = ['cq.lianjia.com']
start_urls = ['cq.lianjia.com']
def __init__(self, **kwargs):
super().__init__(**kwargs)
self.base_url = 'https://cq.lianjia.com'
def start_requests(self):
request_url = 'https://cq.lianjia.com/xiaoqu/'
yield scrapy.Request(url=request_url, callback=self.parse_district_links)
def parse_district_links(self, response):
"""提取地区链接"""
sel = Selector(response)
links = sel.css("div[data-role='ershoufang'] div:first-child a::attr(href)").extract()
for link in links:
url = self.base_url + link
yield scrapy.Request(url=url, callback=self.parse_bizcircle_links)
def parse_bizcircle_links(self, response):
"""提取商圈链接"""
sel = Selector(response)
links = sel.css("div[data-role='ershoufang'] div:nth-child(2) a::attr(href)").extract()
for link in links:
url = self.base_url + link
yield scrapy.Request(url=url, callback=self.parse_village_list, meta={"ref": url})
def parse_village_list(self, response):
"""提取小区链接"""
sel = Selector(response)
links = sel.css(".listContent .xiaoquListItem .img::attr(href)").extract()
for link in links:
yield scrapy.Request(url=link, callback=self.parse_village_detail)
# page
page_data = sel.css(".house-lst-page-box::attr(page-data)").extract_first()
page_data = json.loads(page_data)
if page_data['curPage'] < page_data['totalPage']:
url = response.meta["ref"] + 'pg' + str(page_data['curPage'] + 1)
yield scrapy.Request(url=url, callback=self.parse_village_list, meta=response.meta)
def parse_village_detail(self, response):
"""提取小区详情"""
village_url = response.url
sel = Selector(response)
zone = sel.css('.xiaoquDetailbreadCrumbs .l-txt a::text').extract()
latitude = 0
longitude = 0
try:
html = response.body.decode().replace('\r', '')
local = html[html.find('resblockPosition:'):html.find('resblockName') - 1]
m = re.search('(\d.*\d),(\d.*\d)', local)
longitude = m.group(1)
latitude = m.group(2)
except Exception:
pass
item = LianjiaVillageItem()
item['id'] = village_url.replace(self.base_url + '/xiaoqu/', '').replace('/', '')
item['name'] = sel.css('.detailHeader .detailTitle::text').extract_first()
item['address'] = sel.css('.detailHeader .detailDesc::text').extract_first()
item['latitude'] = latitude
item['longitude'] = longitude
item['zone'] = ','.join(zone)
item['year'] = sel.css('.xiaoquInfo .xiaoquInfoItem:nth-child(1) .xiaoquInfoContent::text').extract_first()
item['build_type'] = sel.css('.xiaoquInfo .xiaoquInfoItem:nth-child(2) .xiaoquInfoContent::text').extract_first()
item['property_costs'] = sel.css('.xiaoquInfo .xiaoquInfoItem:nth-child(3) .xiaoquInfoContent::text').extract_first()
item['property_company'] = sel.css('.xiaoquInfo .xiaoquInfoItem:nth-child(4) .xiaoquInfoContent::text').extract_first()
item['developers'] = sel.css('.xiaoquInfo .xiaoquInfoItem:nth-child(5) .xiaoquInfoContent::text').extract_first()
item['buildings'] = sel.css('.xiaoquInfo .xiaoquInfoItem:nth-child(6) .xiaoquInfoContent::text').extract_first()
item['total_house'] = sel.css('.xiaoquInfo .xiaoquInfoItem:nth-child(7) .xiaoquInfoContent::text').extract_first()
print(item['name'])
yield item
# 小区房源 https://cq.lianjia.com/ershoufang/c3620038190566370/
url = self.base_url + "/ershoufang/c" + item['id'] + "/"
yield scrapy.Request(url=url, callback=self.parse_house_list, meta={"ref": url})
def parse_house_list(self, response):
"""提取房源链接"""
sel = Selector(response)
# 链家有时小区查询不到数据
total = sel.css('.resultDes .total span::text').extract_first()
total = int(total)
if total > 0:
# 提取房源链接
links = sel.css(".sellListContent li .info .title a::attr(href)").extract()
for link in links:
yield scrapy.Request(url=link, callback=self.parse_house_detail)
# 链接分页
page_data = sel.css(".house-lst-page-box::attr(page-data)").extract_first()
page_data = json.loads(page_data)
if page_data['curPage'] == 1 and page_data['totalPage'] > 1:
price = response.url.replace(self.base_url + '/ershoufang/', '')
for x in range(2, page_data['totalPage'] + 1, 1):
url = self.base_url + '/ershoufang/' + 'pg' + str(x) + price
yield scrapy.Request(url=url, callback=self.parse_house_list)
def parse_house_detail(self, response):
"""提取房源信息"""
sel = Selector(response)
item = LianjiaHouseItem()
item['房屋Id'] = response.url.replace(self.base_url + '/ershoufang/', '').replace('.html', '')
item['标题'] = sel.css('.title-wrapper .title .main::text').extract_first()
item['售价'] = sel.css('.overview .content .price .total::text').extract_first()
item['小区'] = sel.css('.overview .content .aroundInfo .communityName a.info::text').extract_first()
item['小区ID'] = sel.css('.overview .content .aroundInfo .communityName a.info::attr(href)').extract_first().replace('/xiaoqu/', '').replace('/', '')
item['房屋户型'] = sel.css('#introduction .base .content ul li:nth-child(1)::text').extract_first()
item['所在楼层'] = sel.css('#introduction .base .content ul li:nth-child(2)::text').extract_first()
item['建筑面积'] = sel.css('#introduction .base .content ul li:nth-child(3)::text').extract_first()
item['户型结构'] = sel.css('#introduction .base .content ul li:nth-child(4)::text').extract_first()
item['套内面积'] = sel.css('#introduction .base .content ul li:nth-child(5)::text').extract_first()
item['建筑类型'] = sel.css('#introduction .base .content ul li:nth-child(6)::text').extract_first()
item['房屋朝向'] = sel.css('#introduction .base .content ul li:nth-child(7)::text').extract_first()
item['建筑结构'] = sel.css('#introduction .base .content ul li:nth-child(8)::text').extract_first()
item['装修情况'] = sel.css('#introduction .base .content ul li:nth-child(9)::text').extract_first()
item['梯户比例'] = sel.css('#introduction .base .content ul li:nth-child(10)::text').extract_first()
item['配备电梯'] = sel.css('#introduction .base .content ul li:nth-child(11)::text').extract_first()
item['产权年限'] = sel.css('#introduction .base .content ul li:nth-child(12)::text').extract_first()
item['挂牌时间'] = sel.css('#introduction .transaction .content ul li:nth-child(1) span:nth-child(2)::text').extract_first()
item['交易权属'] = sel.css('#introduction .transaction .content ul li:nth-child(2) span:nth-child(2)::text').extract_first()
item['上次交易'] = sel.css('#introduction .transaction .content ul li:nth-child(3) span:nth-child(2)::text').extract_first()
item['房屋用途'] = sel.css('#introduction .transaction .content ul li:nth-child(4) span:nth-child(2)::text').extract_first()
item['房屋年限'] = sel.css('#introduction .transaction .content ul li:nth-child(5) span:nth-child(2)::text').extract_first()
item['产权所属'] = sel.css('#introduction .transaction .content ul li:nth-child(6) span:nth-child(2)::text').extract_first()
item['抵押信息'] = sel.css('#introduction .transaction .content ul li:nth-child(7) span:nth-child(2)::attr(title)').extract_first()
item['房本备件'] = sel.css('#introduction .transaction .content ul li:nth-child(8) span:nth-child(2)::text').extract_first()
yield item
完整项目 git地址:https://github.com/oldriverz/house_spider |