粤港澳大湾区7大机场新闻数据的评分系统
# 一、获取网页源代码
# 1.引入所需库
import requests
import re
import time
import csv
# 2.请求头
headers = {'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/69.0.3497.100 Safari/537.36'}
# 3.需要挖取新闻数据的机场
companys = ['广州白云机场', '深圳宝安机场', '惠州平潭机场', '佛山沙堤机场', '珠海金湾机场', '香港国际机场', '澳门国际机场']
# 4.建立挖取某机场新闻的函数,获得该网页的源代码
def baidu(i,company):
num = (i-1)*10
url = 'https://www.baidu.com/s?tn=news&rtt=4&bsst=1&cl=2&wd=' + company + '&medium=0&x_bfe_rqs=03E80&tngroupname=organic_news&newVideo=12&rsv_dl=news_b_pn&pn=' + str(num)
res = requests.get(url, headers=headers).text
# 二、提取信息并清洗数据
# 1.提取源代码中所需的网址、标题、新闻来源和发布日期的正则表达式
p_href = '<h3 class="news-title_1YtI1"><a href="(.*?)"'
href = re.findall(p_href, res, re.S)
p_title = '<h3 class="news-title_1YtI1"><.*?>(.*?)<!--/s-text--></a></h3>'
title = re.findall(p_title, res, re.S)
p_date = '<span class="c-color-gray2 c-font-normal">(.*?)</span>'
date = re.findall(p_date, res, re.S)
p_source = '<span class="c-color-gray c-font-normal c-gap-right">(.*?)</span>'
source = re.findall(p_source, res, re.S)
# 2.清洗标题及发布日期中的干扰信息
for i in range(len(href)):
title[i] = title[i].strip()
title[i] = re.sub('<.*?>', '', title[i])
date[i] = date[i].split(' ')[0]
date[i] = re.sub('年', '-', date[i])
date[i] = re.sub('月', '-', date[i])
date[i] = re.sub('日', '', date[i])
if ('小时' in date[i]) or ('分钟' in date[i]):
date[i] = time.strftime("%Y-%m-%d")
else:
date[i] = date[i]
# 三、对新闻标题和新闻正文进行评分
score = []
keywords1 = ['正常', '涨停', '增加', '超过', '增大', '通过', '恢复', '涨幅', '优质', '首个', '迈入']
keywords2 = ['违约', '诉讼', '下降', '阻碍', '空难', '噪音', '扰民', '黑幕']
for i in range(len(title)):
num = 0
# 1.获取新闻正文
try:
article = requests.get(href[i], headers=headers, timeout=10).text
except:
article = '新闻爬取失败'
# 2.解决新闻正文中可能出现的乱码问题
try:
article = article.encode('ISO-8859-1').decode('utf-8') # 解码为UTF-8
except:
try:
article = article.encode('ISO-8859-1').decode('gbk') # 解码为GBK
except:
article = article # 保持本身的编码
# 3.筛选新闻正文中真正的正文内容,忽略旁边的滚动新闻的内容
p_article = '<p>(.*?)</p>'
article_main = re.findall(p_article, article) # 获取<p>标签里的正文信息,结果是一个列表
article = ''.join(article_main) # 将列表转换成为字符串
for k in keywords1:
if (k in article) or (k in title[i]):
num += 5
score.append(num)
for L in keywords2:
if (L in article) or (L in title[i]):
num -= 5
score.append(num)
# 四、将数据存入csv
# 1. 创建文件对象
f = open('Airport news data.csv','a',encoding='utf-8')
# 2. 基于文件对象构建 csv写入对象
csv_writer = csv.writer(f)
# 3. 构建列表头
csv_writer.writerow(["机场","标题","网址","来源","日期","对新闻标题和内容评分"])
for i in range(len(title)):
csv_writer.writerow([company, title[i], href[i], source[i], date[i], score[i]])
print("yes")
# 4. 关闭文件
f.close()
#5.获取文章内容
# l=open('test.csv','a',encoding='utf-8')
# csv_writer=csv.writer(l)
# csv_writer.writerow([article])
# print("ok!")
# 五、批量爬取7个机场每个机场5页共35页百度网的数据并存入csv
for company in companys:
for i in range(5):
baidu(i, company)
print(company + '第' + str(i+1) + "页爬取成功")
最终结果
附上所有代码
# ===============================
# 粤港澳大湾区7个机场新闻数据的评分系统
# ===============================
# 一、获取网页源代码
# 1.引入所需库
import requests
import re
import time
import csv
# 2.请求头
headers = {'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/69.0.3497.100 Safari/537.36'}
# 3.需要挖取新闻数据的机场
companys = ['广州白云机场', '深圳宝安机场', '惠州平潭机场', '佛山沙堤机场', '珠海金湾机场', '香港国际机场', '澳门国际机场']
# 4.建立挖取某机场新闻的函数,获得该网页的源代码
def baidu(i,company):
num = (i-1)*10
url = 'https://www.baidu.com/s?tn=news&rtt=4&bsst=1&cl=2&wd=' + company + '&medium=0&x_bfe_rqs=03E80&tngroupname=organic_news&newVideo=12&rsv_dl=news_b_pn&pn=' + str(num)
res = requests.get(url, headers=headers).text
# 二、提取信息并清洗数据
# 1.提取源代码中所需的网址、标题、新闻来源和发布日期的正则表达式
p_href = '<h3 class="news-title_1YtI1"><a href="(.*?)"'
href = re.findall(p_href, res, re.S)
p_title = '<h3 class="news-title_1YtI1"><.*?>(.*?)<!--/s-text--></a></h3>'
title = re.findall(p_title, res, re.S)
p_date = '<span class="c-color-gray2 c-font-normal">(.*?)</span>'
date = re.findall(p_date, res, re.S)
p_source = '<span class="c-color-gray c-font-normal c-gap-right">(.*?)</span>'
source = re.findall(p_source, res, re.S)
# 2.清洗标题及发布日期中的干扰信息
for i in range(len(href)):
title[i] = title[i].strip()
title[i] = re.sub('<.*?>', '', title[i])
date[i] = date[i].split(' ')[0]
date[i] = re.sub('年', '-', date[i])
date[i] = re.sub('月', '-', date[i])
date[i] = re.sub('日', '', date[i])
if ('小时' in date[i]) or ('分钟' in date[i]):
date[i] = time.strftime("%Y-%m-%d")
else:
date[i] = date[i]
# 三、对新闻标题和新闻正文进行评分
score = []
keywords1 = ['正常', '涨停', '增加', '超过', '增大', '通过', '恢复', '涨幅', '优质', '首个', '迈入']
keywords2 = ['违约', '诉讼', '下降', '阻碍', '空难', '噪音', '扰民', '黑幕']
for i in range(len(title)):
num = 0
# 1.获取新闻正文
try:
article = requests.get(href[i], headers=headers, timeout=10).text
except:
article = '新闻爬取失败'
# 2.解决新闻正文中可能出现的乱码问题
try:
article = article.encode('ISO-8859-1').decode('utf-8') # 解码为UTF-8
except:
try:
article = article.encode('ISO-8859-1').decode('gbk') # 解码为GBK
except:
article = article # 保持本身的编码
# 3.筛选新闻正文中真正的正文内容,忽略旁边的滚动新闻的内容
p_article = '<p>(.*?)</p>'
article_main = re.findall(p_article, article) # 获取<p>标签里的正文信息,结果是一个列表
article = ''.join(article_main) # 将列表转换成为字符串
for k in keywords1:
if (k in article) or (k in title[i]):
num += 5
score.append(num)
for L in keywords2:
if (L in article) or (L in title[i]):
num -= 5
score.append(num)
# 四、将数据存入csv
# 1. 创建文件对象
f = open('Airport news data.csv','a',encoding='utf-8')
# 2. 基于文件对象构建 csv写入对象
csv_writer = csv.writer(f)
# 3. 构建列表头
csv_writer.writerow(["机场","标题","网址","来源","日期","对新闻标题和内容评分"])
for i in range(len(title)):
csv_writer.writerow([company, title[i], href[i], source[i], date[i], score[i]])
print("yes")
# 4. 关闭文件
f.close()
#5.获取文章内容
# l=open('test.csv','a',encoding='utf-8')
# csv_writer=csv.writer(l)
# csv_writer.writerow([article])
# print("ok!")
# 五、批量爬取7个机场每个机场5页共35页百度网的数据并存入csv
for company in companys:
for i in range(5):
baidu(i, company)
print(company + '第' + str(i+1) + "页爬取成功")