Python爬虫自动化获取华图和粉笔网站的错题(推荐)


Posted in Python onJanuary 08, 2021

这篇博客对于考公人或者其他用华图或者粉笔做题的人比较友好,通过输入网址可以自动化获取华图以及粉笔练习的错题。

粉笔网站

我们从做过的题目组中获取错题

Python爬虫自动化获取华图和粉笔网站的错题(推荐)
Python爬虫自动化获取华图和粉笔网站的错题(推荐)

打开某一次做题组,我们首先进行抓包看看数据在哪里

Python爬虫自动化获取华图和粉笔网站的错题(推荐)

我们发现现在数据已经被隐藏,事实上数据在这两个包中:
https://tiku.fenbi.com/api/xingce/questions
https://tiku.fenbi.com/api/xingce/solutions
一个为题目的一个为解析的。此url要通过传入一个题目组参数才能获取到当前题目数据,而题目组参数在这个包中

Python爬虫自动化获取华图和粉笔网站的错题(推荐)

以网址的倒数第二个数字串有关

Python爬虫自动化获取华图和粉笔网站的错题(推荐)

url的规则为'https://tiku.fenbi.com/api/xingce/exercises/'+str(id_)+'?app=web&kav=12&version=3.0.0.0',id_即为下划线数字
通过请求这个包获取到参数然后通过参数请求上面两个包(

https://tiku.fenbi.com/api/xingce/questions
https://tiku.fenbi.com/api/xingce/solutions

)即可获取到题目数据,而且自己的答案在也在https://tiku.fenbi.com/api/xingce/exercises/'+str(id_)+'?app=web&kav=12&version=3.0.0.0这个包中。

不过粉笔的题目数据有些是图片,而且图片在题目中,选项中,这里以word文档存储操作docx库有些吃力,于是我想到了直接构造HTML代码,然后通过pdfkit转为pdf(具体如何下载可以参考百度,要下载wkhtmltopdf.exe)即可变为错题集在平板或者其他设备中看。
(请求时一定要携带完整的headers,否则很可能获取不到数据)

具体操作看代码解析

###此函数用于解析题目和每道题的答案
def jiexi(liebiao):
 new = []
 timu_last = []
 for each in liebiao:
  new.append(re.sub(r'flag=\\"tex\\" ','',each))
 for each in new:
  timu_last.append(re.sub(r'\\','',each))
 return timu_last
###此函数用于解析选项
def xuanxiang(liebiao):
 xuanxiang_v2 = []
 xuanxiang_v3 = []
 for each in liebiao:
  a = re.sub('<p>','',each)
  a = re.sub('</p>','',a)
  xuanxiang_v2.append(a)
 for each in xuanxiang_v2:
  each = each+'</p>'
  xuanxiang_v3.append(each)
 return xuanxiang_v3
import requests
import re
import pdfkit
import os
url = str(input("请输入练习的网址:"))
###获取本节练习id
id_ = re.findall(r'https://www.fenbi.com/spa/tiku.*?/xingce/xingce/(.*?)/',url,re.S)[0]
mid_url = 'https://tiku.fenbi.com/api/xingce/exercises/'+str(id_)+'?app=web&kav=12&version=3.0.0.0'
headers = {
#####完整的headers,自己添加
}
response = requests.get(url=mid_url,headers=headers)
response.encoding = 'utf-8'
page_text = response.text
###获取题目组参数
id_list = re.findall('\"questionIds\"\:\[(.*?)\]\,',page_text,re.S)
###获取自己的答案
your_answer = re.findall(r'"answer":{"choice":"(.*?)",',page_text,re.S)
###此练习名称
name = re.findall(r'"name":"(.*?)",',page_text,re.S)[0]
###真正存储数据的包
timu_url = 'https://tiku.fenbi.com/api/xingce/questions'
params = {
 'ids': id_list
}
response = requests.get(url=timu_url,headers=headers,params=params)
response.encoding = 'utf-8'
page_text = response.text
###获取正确答案
true_answer = re.findall('"correctAnswer":{"choice":"(.*?)"',page_text,re.S)
###真正存储数据的包
solution_url = 'https://tiku.fenbi.com/api/xingce/solutions'
response = requests.get(url=solution_url,headers=headers,params=params)
response.encoding = 'utf-8'
page_text = response.text
###获取解析
solution_list = re.findall(r'"solution":"(.*?)","userAnswer"',page_text,re.S)
solution_last = jiexi(solution_list)
cailiao = []
timu = []
###获取单选题题目和复合题的题目
for each in response.json():
 timu.append(each['content'])
 try:
  cailiao.append(each['material']['content'])
 except:
  cailiao.append('none')
###获取选项信息
A_option = re.findall('\"options\"\:\[\"(.*?)\"\,\".*?\"\,\".*?\"\,\".*?\"\]',page_text,re.S)
B_option = re.findall('\"options\"\:\[\".*?\"\,\"(.*?)\"\,\".*?\"\,\".*?\"\]',page_text,re.S)
C_option = re.findall('\"options\"\:\[\".*?\"\,\".*?\"\,\"(.*?)\"\,\".*?\"\]',page_text,re.S)
D_option = re.findall('\"options\"\:\[\".*?\"\,\".*?\"\,\".*?\"\,\"(.*?)\"\]',page_text,re.S)
A_option = xuanxiang(A_option)
B_option = xuanxiang(B_option)
C_option = xuanxiang(C_option)
D_option = xuanxiang(D_option)
A_option = jiexi(A_option)
B_option = jiexi(B_option)
C_option = jiexi(C_option)
D_option = jiexi(D_option)
###构造HTML代码
count = 0
all_content = "<!DOCTYPE html>\n<meta charset='utf-8'>\n<html>"
for each in true_answer:
 if each != your_answer[count]:
  ###处理复合题
  if cailiao[count] != 'none' and cailiao[count] not in all_content:
   all_content += cailiao[count]
  all_content += str(count+1)
  all_content += '、'
  all_content += timu[count][3:]
  all_content += 'A、'
  all_content += A_option[count]
  all_content += 'B、'
  all_content += B_option[count]
  all_content += 'C、'
  all_content += C_option[count]
  all_content += 'D、'
  all_content += D_option[count]
  all_content += '<br>'
 count += 1
count = 0
all_content += '<br><br><br><br><br><br><br><br><br>'
for each in true_answer:
 if each != your_answer[count]:
  temp = '第'+str(count+1)+'题的正确答案为'
  all_content += temp
  if true_answer[count]=='0':
   all_content += 'A'
  elif true_answer[count]=='1':
   all_content += 'B'
  elif true_answer[count]=='2':
   all_content += 'C'
  elif true_answer[count]=='3':
   all_content += 'D'
  all_content += solution_last[count]
  all_content += '<br>'
 count += 1
all_content += '</html>'
path_name = name + '.html'
###保存为HTML文件
with open(path_name,'w',encoding='utf-8') as fp:
 fp.write(all_content)
confg = pdfkit.configuration(wkhtmltopdf=r'wkhtmltopdf.exe保存的路径')
pdfkit.from_url(path_name, name+'.pdf',configuration=confg)###把HTML文件转为pdf
print('错题PDF保存成功')
###删除HTML文件
os.remove(path_name)

华图网站

Python爬虫自动化获取华图和粉笔网站的错题(推荐)

也是答题记录中自己做过的题目
华图网站稍微不一样,他的数据直接抓包就可看到

Python爬虫自动化获取华图和粉笔网站的错题(推荐)

通过请求这个包即可获取到数据,接下来就是解析的事情了,这次我用word文档进行存储,如果觉得不方便也可以像上文一样构造HTML

##导包
import requests
import lxml.etree
import re
import time
import os
from docx import Document
from docx.shared import Inches
from docx.shared import Pt
from docx.shared import Inches
from docx.oxml.ns import qn
from docx.enum.text import WD_ALIGN_PARAGRAPH
url = str(input("请输入练习的网址:"))
headers={
###完整的headers,否则获取不到数据
}
response = requests.get(url = url,headers = headers)
response.encoding='utf-8'
reptext = response.text
tree = lxml.etree.HTML(reptext) #解析网站获取源码

dirName="考公图片"
if not os.path.exists(dirName):
 os.mkdir(dirName) #网站图片保存路径
 
jiexi = re.findall(r'<div class="jiexi-item-title">解析.*?。</div>.*?</div>', reptext,re.S) #获取题目解析

imgg = []
for each in jiexi:
 imgg.append(re.findall(r'<img src="(.*?)".*?>', each)) #获取解析里的图片URL
 
imgt = []
for each in imgg:
 if each == []:
  imgt.append([1])
 else:
  imgt.append(each) #把解析里图片URL美化整理一下
  
jiexilast = []
for qq in jiexi:
 jiexilast.append(re.sub(r'<[^>]+>', '', qq)) #美化题目解析
 
corrected = re.findall(r'<span class="g-right-answer-color">[a-zA-Z]{1,4}</span>', reptext) #获取正确答案
correct = []
for ee in corrected:
 correct.append(re.sub(r'<[^>]+>', '', ee)) #美化正确答案
 
yoursed = re.findall(r'<span class="yellowWord">[a-zA-Z]{1,4}</span>', reptext) #获取自己的答案
yours = []
for ee in yoursed:
 yours.append(re.sub(r'<[^>]+>', '', ee)) #美化自己的答案
 
timuleixing = re.findall(r'<span class="greenWord">(.*?)</span>.*?</div>',reptext,re.S) #获取题目类型

find1 = re.findall(r'<span class="greenWord">.*?</span>(.*?)</div>',reptext,re.S)
for each in find1:
 re.sub(r'<.*?>','',each)
find5 = [] #最终的题目
for each in find1:
 find5.append(re.sub(r'<[^>]+>', '', each))
 
img = []
for each in find1:
 img.append(re.findall(r'<img src="(.*?)".*?>', each))
imgx = []
for each in img:
 if each == []:
  imgx.append([1])
 else:
  imgx.append(each) #最终版题目图片URL
  

v = tree.xpath('//div[@class="exercise-main-title"]//text()') #本次题目类型

try:
 ###这是既有复合题也有单选题的
 fuheti = re.findall(r'<!--复合题-->(.*?)<div class="exercise-main-topics"',reptext,re.S)[0].split('<!--复合题-->')
except:
 try:
  ###这是只有复合题或者复合题在最后几题的
  fuheti = re.findall(r'<!--复合题-->(.*?)<!-- 纠错的弹窗 -->',reptext,re.S)[0].split('<!--复合题-->')
 except:
  pass
count = 0

###导入标题
document = Document()
p = document.add_paragraph()
p.paragraph_format.alignment = WD_ALIGN_PARAGRAPH.CENTER
run = p.add_run(v[0][5:-3])
run.font.size = Pt(14)
run.font.name=u'宋体'
r = run._element
r.rPr.rFonts.set(qn('w:eastAsia'),u'宋体')
choose = []

###处理题目选项
axuanxiang = []
bxuanxiang = []
cxuanxiang = []
dxuanxiang = []
xuanxiang = re.findall(r'<div class="main-topic-choices">(.*?)<div class="main-topic-letters clearfix pl14">',reptext,re.S)
for everything in xuanxiang:
 try: ##处理只有两个选项
  axuanxiang.append(re.sub("<.*?>","",re.findall(r'<div.*?class.*?main-topic-choice.*?>(A.*?)</div>',everything,re.S)[0]))
 except:
  axuanxiang.append('--')
 try:
  bxuanxiang.append(re.sub("<.*?>","",re.findall(r'<div.*?class.*?main-topic-choice.*?>(B.*?)</div>',everything,re.S)[0]))
 except:
  bxuanxiang.append('--')
 try:
  cxuanxiang.append(re.sub("<.*?>","",re.findall(r'<div.*?class.*?main-topic-choice.*?>(C.*?)</div>',everything,re.S)[0]))
 except:
  cxuanxiang.append('--')
 try:
  dxuanxiang.append(re.sub("<.*?>","",re.findall(r'<div.*?class.*?main-topic-choice.*?>(D.*?)</div>',everything,re.S)[0]))
 except:
  dxuanxiang.append('--')
  

 
for every in correct:
 if every != yours[count]:
  ###处理复合题题目
  try:
   for eacy in fuheti:
    if find5[count] in eacy:
     fuheti_URL = re.findall(r'<img src="(.*?)".*?>',re.findall(r'.*?<p>(.*?)</p>',eacy,re.S)[0],re.S)
     fuheti_last = re.sub(r'<.*?>','',re.findall(r'.*?<p>(.*?)</p>',eacy,re.S)[0])
     fuheti_last = re.sub(r'\xa0\xa0\xa0\xa0\xa0\xa0\xa0','\n',fuheti_last)
     if fuheti_last not in choose:
      p = document.add_paragraph()
      run = p.add_run(fuheti_last)
      run.font.size = Pt(14)
      run.font.name=u'宋体'
      r = run._element
      r.rPr.rFonts.set(qn('w:eastAsia'),u'宋体')
      headers ={
     'Use-Agent':'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/81.0.4044.122 Safari/537.36'
       }
      for eacu in fuheti_URL:
       img_data = requests.get(url = eacu,headers = headers).content
       img_path = dirName+'/'+'tupian'+'.jpg'
       with open(img_path,'wb') as fp:
        fp.write(img_data)
        print("保存成功")
       document.add_picture(img_path, width=Inches(5))
      choose.append(fuheti_last)
  except:
   pass
  
  ###导入单选题题目
  p = document.add_paragraph()
  run = p.add_run(str(count+1)+"、"+timuleixing[count]+find5[count][3:])
  run.font.size = Pt(14)
  run.font.name=u'宋体'
  r = run._element
  r.rPr.rFonts.set(qn('w:eastAsia'),u'宋体')
  url = imgx[count][0]
  headers ={
   'Use-Agent':'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/81.0.4044.122 Safari/537.36'
  }
  try:
   img_data = requests.get(url = url,headers = headers).content
   img_path = dirName+'/'+'tupian'+'.jpg'
   with open(img_path,'wb') as fp:
    fp.write(img_data)
    print("保存成功")
   document.add_picture(img_path, width=Inches(5))
   count+=1
  except:
   count+=1
   
  ###导入选项
  p = document.add_paragraph()
  run = p.add_run(axuanxiang[count-1])
  run.font.size = Pt(14)
  run.font.name=u'宋体'
  r = run._element
  r.rPr.rFonts.set(qn('w:eastAsia'),u'宋体')
  p = document.add_paragraph()
  run = p.add_run(bxuanxiang[count-1])
  run.font.size = Pt(14)
  run.font.name=u'宋体'
  r = run._element
  r.rPr.rFonts.set(qn('w:eastAsia'),u'宋体')
  p = document.add_paragraph()
  run = p.add_run(cxuanxiang[count-1])
  run.font.size = Pt(14)
  run.font.name=u'宋体'
  r = run._element
  r.rPr.rFonts.set(qn('w:eastAsia'),u'宋体')
  p = document.add_paragraph()
  run = p.add_run(dxuanxiang[count-1])
  run.font.size = Pt(14)
  run.font.name=u'宋体'
  r = run._element
  r.rPr.rFonts.set(qn('w:eastAsia'),u'宋体')
  p = document.add_paragraph()
  run = p.add_run("\n")
  run.font.size = Pt(14)
  run.font.name=u'宋体'
  r = run._element
  r.rPr.rFonts.set(qn('w:eastAsia'),u'宋体')
  
 else:
  count+=1

###美化界面
p = document.add_paragraph()
run = p.add_run("\n\n\n\n\n")
run.font.size = Pt(14)
run.font.name=u'宋体'
r = run._element
r.rPr.rFonts.set(qn('w:eastAsia'),u'宋体')

###美化解析
counting = 0
jiexilast2 = []
for ok in jiexilast:
 jiexilast2.append(re.sub(r'\n\t\t',':',ok))
for every in correct:
 if every != yours[counting]:
  ###导入解析和答案
  p = document.add_paragraph()
  run = p.add_run(str(counting+1)+"、"+"正确答案为:"+correct[counting]+"\n"+jiexilast2[counting])
  run.font.size = Pt(14)
  run.font.name=u'宋体'
  r = run._element
  r.rPr.rFonts.set(qn('w:eastAsia'),u'宋体')
  url = imgt[counting][0]
  headers ={
   'Use-Agent':'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/81.0.4044.122 Safari/537.36'
  }
  try:
   img_data = requests.get(url = url,headers = headers).content
   img_path = dirName+'/'+'tupian'+'.jpg'
   with open(img_path,'wb') as fp:
    fp.write(img_data)
    print("保存成功")
   document.add_picture(img_path, width=Inches(5))
   print("写入成功")
   counting+=1
  except:
   counting+=1
 else:
  counting+=1
###保存文档
document.save(v[0][5:-3]+'.docx')
print(v[0][5:-3]+'保存成功!')

总结

粉笔和华图错题爬虫主要区别是华图获取数据简单,解析操作繁琐;粉笔的数据隐秘,解析起来可以用json,比较方便。

到此这篇关于Python爬虫自动化获取华图和粉笔网站的错题的文章就介绍到这了,更多相关Python爬虫获取网站错题内容请搜索三水点靠木以前的文章或继续浏览下面的相关文章希望大家以后多多支持三水点靠木!

Python 相关文章推荐
Python中函数的参数传递与可变长参数介绍
Jun 30 Python
使用Python发送各种形式的邮件的方法汇总
Nov 09 Python
Python3生成手写体数字方法
Jan 30 Python
Python实现使用卷积提取图片轮廓功能示例
May 12 Python
Python lxml解析HTML并用xpath获取元素的方法
Jan 02 Python
教你如何编写、保存与运行Python程序的方法
Jul 12 Python
Pyinstaller 打包exe教程及问题解决
Aug 16 Python
python3 实现的对象与json相互转换操作示例
Aug 17 Python
pycharm实现在子类中添加一个父类没有的属性
Mar 12 Python
keras处理欠拟合和过拟合的实例讲解
May 25 Python
matplotlib 三维图表绘制方法简介
Sep 20 Python
Pandas之缺失数据的实现
Jan 06 Python
tensorflow与numpy的版本兼容性问题的解决
Jan 08 #Python
matplotlib自定义鼠标光标坐标格式的实现
Jan 08 #Python
selenium设置浏览器为headless无头模式(Chrome和Firefox)
Jan 08 #Python
python画图时设置分辨率和画布大小的实现(plt.figure())
Jan 08 #Python
python使用matplotlib的savefig保存时图片保存不完整的问题
Jan 08 #Python
Numpy中的数组搜索中np.where方法详细介绍
Jan 08 #Python
python 窃取摄像头照片的实现示例
Jan 08 #Python
You might like
php 防止单引号,双引号在接受页面转义
2008/07/10 PHP
解析crontab php自动运行的方法
2013/06/24 PHP
PHP加密解密函数详解
2015/10/28 PHP
利用PHP扩展Xhprof分析项目性能实践教程
2018/09/05 PHP
PHP进阶学习之类的自动加载机制原理分析
2019/06/18 PHP
javascript 动态修改样式和层叠样式表代码
2010/04/27 Javascript
Jquery练习之表单验证实现代码
2010/12/14 Javascript
Dreamweaver jQuery智能提示插件,支持版本提示,支持1.6api
2011/07/31 Javascript
JSCode all of Brower 全局屏蔽网页右键功能 具体实现
2013/06/05 Javascript
jQuery 获取浏览器所在的IP地址的小例子
2013/11/08 Javascript
javascript实现的HashMap类代码
2014/06/27 Javascript
js实现分享到随页面滚动而滑动效果的方法
2015/04/10 Javascript
动态加载jQuery的方法
2015/06/16 Javascript
浅谈JS中的bind方法与函数柯里化
2016/08/10 Javascript
jQuery实现表格冻结顶栏效果
2017/08/20 jQuery
vue toggle做一个点击切换class(实例讲解)
2018/03/13 Javascript
js中this的指向问题归纳总结
2018/11/28 Javascript
一次微信小程序内地图的使用实战记录
2019/09/09 Javascript
Vue中axios拦截器如何单独配置token
2019/12/27 Javascript
Python中的连接符(+、+=)示例详解
2017/01/13 Python
Python实现向服务器请求压缩数据及解压缩数据的方法示例
2017/06/09 Python
python仿evething的文件搜索器实例代码
2019/05/13 Python
django框架自定义模板标签(template tag)操作示例
2019/06/24 Python
使用Python发现隐藏的wifi
2020/03/04 Python
维珍澳洲航空官网:Virgin Australia
2017/09/08 全球购物
UNOde50美国官网:西班牙珠宝品牌
2020/08/15 全球购物
PHP开发的一般流程
2013/08/13 面试题
如何利用cmp命令比较文件
2016/04/11 面试题
如何通过 CSS 写出火焰效果
2021/03/24 HTML / CSS
黄金搭档广告词
2014/03/21 职场文书
材料物理专业求职信
2014/09/01 职场文书
2014年工作总结与下年工作计划
2014/11/27 职场文书
2015新学期家长寄语
2015/02/26 职场文书
创建文明城市倡议书
2015/04/28 职场文书
HTML基础详解(下)
2021/10/16 HTML / CSS
使用python生成大量数据写入es数据库并查询操作(2)
2022/09/23 Python