标签:
# -*- coding: utf-8 -*- import urllib.request import ssl import json import xlwt context = ssl._create_unverified_context() title=[‘女装‘,‘鞋包‘,‘男士‘,‘运动‘,‘饰品‘,‘美妆‘,‘母婴‘,‘居家‘,‘国际‘,‘生活‘] wb = xlwt.Workbook() class spider: def url_name(self): wb = xlwt.Workbook() for i in range(len(title)): list_name = [] list_img = [] print(i+1) ws = wb.add_sheet(title[i]) url = ‘http://www.vip.com/index-ajax.php?act=getSellingBrandListV5&warehouse=VIP_NH&areaCode=104104&channelId=0&pagecode=b&sortType=1&province_name=%E5%B9%BF%E4%B8%9C&city_name=%E5%B9%BF%E5%B7%9E%E5%B8%82&preview=&sell_time_from=&time_from=&ids=‘+str(i+1) url_data = urllib.request.urlopen(url).read().decode("utf-8") print(url_data) jsDict = json.loads(url_data) print(jsDict) jsdata = jsDict[‘data‘] jsfloor = jsdata[‘floors‘] jsfirst = jsfloor[str(i+1)] jsitems = jsfirst[‘items‘] for each in jsitems: list_img.append(each[‘mobile_image_one‘]) list_name.append(each[‘name‘]) print(len(list_img)) print(len(list_name)) print(list_name) print(list_img) for each in range(len(list_name)): ws.write(each, 0, list_name[each]) ws.write(each, 1, list_img[each]) x=0 for j in list_img: urllib.request.urlretrieve(j, ‘D:\\weipinhui\\monning_1\\jingxuan_pic\\‘+str(i+1)+‘\\%s.jpg‘ % x) x = x + 1 wb.save(‘D:\\weipinhui\\monning_1\\jingxuan_name_url\\jingxuan_name_url.xls‘) def shouye(self): wb = xlwt.Workbook() ws = wb.add_sheet(‘首页‘) list_name_sy=[] list_img_sy=[] url = ‘http://pcapi.vip.com/ads/index.php?callback=shopAds&type=ADSEC56K%2CADSIR7IX%2CADSX7W3G%2CADSNNLS7%2CADS7JI3F%2CADS2B669%2CADSITG64%2CADS45AV4%2CADS44T33&warehouse=VIP_NH&areaid=104104&preview=0&date_from=&time_from=&user_class=&channelId=0‘ url_data = urllib.request.urlopen(url).read().decode("utf-8") url_data = url_data.replace(‘shopAds(‘, ‘‘) url_data = url_data.replace(‘)‘, ‘‘) jsDict = json.loads(url_data) print(jsDict) jsdata = jsDict[‘ADADSEC56K‘] jsdatas = jsdata[‘items‘] for each in jsdatas: list_name_sy.append(each[‘name‘]) list_img_sy.append(each[‘img‘]) print(list_img_sy) print(list_name_sy) x = 0 for each in range(len(list_name_sy)): ws.write(each, 0, list_name_sy[each]) ws.write(each, 1, list_img_sy[each]) for j in list_img_sy: urllib.request.urlretrieve(j, ‘D:\\weipinhui\\monning_1\\shouye_pic\\%s.jpg‘ % x) x = x + 1 wb.save(‘D:\\weipinhui\\monning_1\\shouye_name_url\\shouye_name_url.xls‘) content=spider() content.url_name() content.shouye()
标签:
原文地址:http://www.cnblogs.com/caicaihong/p/5925666.html