Hadoop综合大作业

Hadoop综合大作业 要求：

1. 用Hive对爬虫大作业产生的文本文件（或者英文词频统计下载的英文长篇小说）词频统计。

2. 用Hive对爬虫大作业产生的csv文件进行数据分析

1. 用Hive对爬虫大作业产生的文本文件

这里的具体操作步骤如下：

• 将网页上的歌词段落爬取下来
• 通过jieba分词后将结果用txt文件保存，
• 将txt文件放入Hadoop分布式文件系统
• 使用hive将文件作为表数据导入
• 使用hive查询统计歌词中单词的出现次数

首先，Python爬虫程序代码如下：

import jieba
import requests
from bs4 import BeautifulSoup

lyrics = ''
headers = {
    'User-Agent': 'User-Agent:*/*'
}

resp = requests.get('http://www.juzimi.com/writer/%E6%96%B9%E6%96%87%E5%B1%B1', headers=headers)
resp.encoding = 'UTF-8'
print(resp.status_code)
soup = BeautifulSoup(resp.text, 'html.parser')

page_url = 'http://www.juzimi.com/writer/%E6%96%B9%E6%96%87%E5%B1%B1?page={}'
page_last = soup.select('.pager-last')
if len(page_last) > 0:
    page_last = page_last[0].text

for i in range(0, int(page_last)):
    print(i)
    resp = requests.get(page_url.format(i), headers=headers)
    resp.encoding = 'UTF-8'
    soup = BeautifulSoup(resp.text, 'html.parser')
    for a in soup.select('.xlistju'):
        lyrics += a.text + ' '

# 保留爬取的句子
with open('lyrics.txt', 'a+', encoding='UTF-8') as lyricFile:
    lyricFile.write(lyrics)

# 加载标点符号并去除歌词中的标点
with open('punctuation.txt', 'r', encoding='UTF-8') as punctuationFile:
    for punctuation in punctuationFile.readlines():
        lyrics = lyrics.replace(punctuation[0], ' ')

# 加载无意义词汇
with open('meaningless.txt', 'r', encoding='UTF-8') as meaninglessFile:
    mLessSet = set(meaninglessFile.read().split('\n'))
mLessSet.add(' ')

# 加载保留字
with open('reservedWord.txt', 'r', encoding='UTF-8') as reservedWordFile:
    reservedWordSet = set(reservedWordFile.read().split('\n'))
    for reservedWord in reservedWordSet:
        jieba.add_word(reservedWord)

keywordList = list(jieba.cut(lyrics))
keywordSet = set(keywordList) - mLessSet  # 将无意义词从词语集合中删除
keywordDict = {}

# 统计出词频字典
for word in keywordSet:
    keywordDict[word] = keywordList.count(word)

# 对词频进行排序
keywordListSorted = list(keywordDict.items())
keywordListSorted.sort(key=lambda e: e[1], reverse=True)
# 将所有词频写出到txt
for topWordTup in keywordListSorted:
    print(topWordTup)
    with open('word.txt', 'a+', encoding='UTF-8') as wordFile:
        for i in range(0, topWordTup[1]):
            wordFile.write(topWordTup[0]+'\n')

现在将word.txt放入HDFS中并用hive查询统计，命令如下：

hdfs dfs -mkdir temp
hdfs dfs -put news.csv temp
hive

hive>
	create database db_temp;
	use db_temp;
	create table tb_word(word string);
	load data inpath '/user/hadoop/temp/word.txt' into table tb_word;
	select word, count(1) as num from tb_word group by word order by num desc limit 50;

以上的运行结果截图如下：

2. 用Hive对爬虫大作业产生的csv文件进行数据分析

我这里选择了爬取校园新闻并生产csv文件来分析，首先编写爬虫，主要代码如下：

import requests
from bs4 import BeautifulSoup
from datetime import datetime
import re
import pandas

news_list = []

def crawlOnePageSchoolNews(page_url):
    res0 = requests.get(page_url)
    res0.encoding = 'UTF-8'
    soup0 = BeautifulSoup(res0.text, 'html.parser')
    news = soup0.select('.news-list > li')

    for n in news:
        # print(n)
        print('**' * 5 + '列表页信息' + '**' * 10)
        print('新闻链接：' + n.a.attrs['href'])
        print('新闻标题：' + n.select('.news-list-title')[0].text)
        print('新闻描述：' + n.a.select('.news-list-description')[0].text)
        print('新闻时间：' + n.a.select('.news-list-info > span')[0].text)
        print('新闻来源：' + n.a.select('.news-list-info > span')[1].text)
        news_list.append(getNewDetail(n.a.attrs['href']))
    return news_list

def getNewDetail(href):
    print('**' * 5 + '详情页信息' + '**' * 10)
    print(href)
    res1 = requests.get(href)
    res1.encoding = 'UTF-8'
    soup1 = BeautifulSoup(res1.text, 'html.parser')
    news = {}
    if soup1.select('#content'):
        news_content = soup1.select('#content')[0].text
        news['content'] = news_content.replace('\n', ' ').replace('\r', ' ').replace(',', '·')
        print(news_content)  # 文章内容
    else:
        news['content'] = ''
    if soup1.select('.show-info'):  # 防止之前网页没有show_info
        news_info = soup1.select('.show-info')[0].text
    else:
        return news
    info_list = ['来源', '发布时间', '点击', '作者', '审核', '摄影']  # 需要解析的字段
    news_info_set = set(news_info.split('\xa0')) - {' ', ''}  # 网页中的 获取后会解析成\xa0，所以可以使用\xa0作为分隔符
    # 循环打印文章信息
    for n_i in news_info_set:
        for info_flag in info_list:
            if n_i.find(info_flag) != -1:  # 因为时间的冒号采用了英文符所以要进行判断
                if info_flag == '发布时间':
                    # 将发布时间字符串转为datetime格式,方便日后存储到数据库
                    release_time = datetime.strptime(n_i[n_i.index(':') + 1:], '%Y-%m-%d %H:%M:%S ')
                    news[info_flag] = release_time
                    print(info_flag + ':', release_time)
                elif info_flag == '点击':  # 点击次数是通过文章id访问php后使用js写入，所以这里单独处理
                    news[info_flag] = getClickCount(href)
                else:
                    news[info_flag] = n_i[n_i.index('：') + 1:].replace(',', '·')
                    print(info_flag + ':' + n_i[n_i.index('：') + 1:])
    print('————' * 40)
    return news

def getClickCount(news_url):
    click_num_url = 'http://oa.gzcc.cn/api.php?op=count&id={}&modelid=80'
    click_num_url = click_num_url.format(re.search('_(.*)/(.*).html', news_url).group(2))
    res2 = requests.get(click_num_url)
    res2.encoding = 'UTF-8'
    click_num = re.search("\$\('#hits'\).html\('(\d*)'\)", res2.text).group(1)
    print('点击:' + click_num)
    return click_num

print(crawlOnePageSchoolNews('http://news.gzcc.cn/html/xiaoyuanxinwen/'))

pageURL = 'http://news.gzcc.cn/html/xiaoyuanxinwen/{}.html'
res = requests.get('http://news.gzcc.cn/html/xiaoyuanxinwen/')
res.encoding = 'UTF-8'
soup = BeautifulSoup(res.text, 'html.parser')
newsSum = int(re.search('(\d*)条', soup.select('a.a1')[0].text).group(1))
if newsSum % 10:
    pageSum = int(newsSum / 10) + 1
else:
    pageSum = int(newsSum / 10)

for i in range(2, pageSum + 1):
    crawlOnePageSchoolNews(pageURL.format(i))

# with open('news.txt', 'w') as file:
#     file.write()

dit = pandas.DataFrame(news_list)
dit.to_csv('news.csv')
print(dit)

因为csv是用逗号分隔，而文章内容有逗号和换行符容易造成影响，所以在爬取数据时做了相应处理，将换行逗号等使用其他代替。爬取后将文件放入HDFS系统，并将第一行的数据删除，这里使用insert语句覆盖原先导入的表即可，然后通过hive查询做出相应操作分析文章作者在什么时候发表的量比较多。

hdfs dfs -put  news.csv temp/
hive

hive>
create table tb_news(id string, content string, author string, publish timestamp, verify string, photo string, source string, click int)row format delimited fields terminated by ',';

load data inpath '/user/hadoop/temp/news.csv' overwrite into table tb_news;

insert overwrite table tb_news select * from tb_news where content != 'content';

select time_publish, count(1) as num from (select hour(publish) as time_publish from tb_news) tb_time group by time_publish order by num desc;

根据以上截图的结果可以看出，小编在发布时间大部分都是在0时，我只能说，熬夜不好