Python Scrapy 爬虫框架爬取推特信息及数据持久化

最近要做一个国内外新冠疫情的热点信息的收集系统,所以,需要爬取推特上的一些数据,然后做数据分类及情绪分析。作为一名合格的程序员,我们要有「拿来主义精神」,借助别人的轮子来实现自己的项目,而不是从头搭建。

一、爬虫框架Scrapy

Scrapy 是用Python实现一个为爬取网站数据、提取结构性数据而编写的应用框架。专业的事情交给专业的框架来做,所以,本项目我们确定使用 Scrapy 框架来进行数据爬取。如果对 Scrapy 还不熟悉,可以看我之前写的这篇博文帮你快速上手,Python Scrapy爬虫框架学习

二、寻找开源项目

在开始一个项目之前,避免重复造轮子,所以通过关键词 「Scrapy」,「Twitter」在 GitHub上搜索是否有现成的开源项目。

file

通过搜索,我们发现有很多符合条件的开源项目,那么如何选择这些项目呢?有三个条件,第一是Star数,Star数多说明项目质量应该不错得到了大家的认可,第二是,更新时间,说明这个项目一直在维护,第三是,文档是否完整,通过文档我们可以快速使用这个开源项目。所以,通过以上三个条件,我们看了下排在第一个的开源项目很不错,star数颇高,最近更新时间在几个月前,而且文档很详细,因此我们就用这个项目做二次开发,项目GitHub地址:jonbakerfish/TweetScraper

三、本地安装及调试

1、拉取项目

It requires Scrapy and PyMongo (Also install MongoDB if you want to save the data to database). Setting up:

$ git clone https://github.com/jonbakerfish/TweetScraper.git
$ cd TweetScraper/
$ pip install -r requirements.txt  #add '--user' if you are not root
$ scrapy list
$ #If the output is 'TweetScraper', then you are ready to go.

2、数据持久化

通过阅读文档,我们发现该项目有三种持久化数据的方式,第一种是保存在文件中,第二种是保存在Mongo中,第三种是保存在MySQL数据库中。因为我们抓取的数据需要做后期的分析,所以,需要将数据保存在MySQL中。

抓取到的数据默认是以Json格式保存在磁盘 ./Data/tweet/ 中的,所以,需要修改配置文件 TweetScraper/settings.py

ITEM_PIPELINES = {
    # 'TweetScraper.pipelines.SaveToFilePipeline':100,
    #'TweetScraper.pipelines.SaveToMongoPipeline':100, # replace `SaveToFilePipeline` with this to use MongoDB
    'TweetScraper.pipelines.SavetoMySQLPipeline':100, # replace `SaveToFilePipeline` with this to use MySQL
}

#settings for mysql
MYSQL_SERVER = "18.126.219.16"
MYSQL_DB     = "scraper"
MYSQL_TABLE  = "tweets" # the table will be created automatically
MYSQL_USER   = "root"        # MySQL user to use (should have INSERT access granted to the Database/Table
MYSQL_PWD    = "admin123456"        # MySQL user's password

3、测试

进入到项目的根目录下,运行以下命令:

 # 进入到项目目录
# cd  /work/Code/scraper/TweetScraper 
 scrapy crawl TweetScraper -a query="Novel coronavirus,#COVID-19"
注意,抓取Twitter的数据需要科学上网或者服务器部署在国外,我使用的是国外的服务器。
[root@cs TweetScraper]#  scrapy crawl TweetScraper -a query="Novel coronavirus,#COVID-19"
2020-04-16 19:22:40 [scrapy.utils.log] INFO: Scrapy 2.0.1 started (bot: TweetScraper)
2020-04-16 19:22:40 [scrapy.utils.log] INFO: Versions: lxml 4.2.1.0, libxml2 2.9.8, cssselect 1.1.0, parsel 1.5.2, w3lib 1.21.0, Twisted 20.3.0, Python 3.6.5 |Anaconda, Inc.| (default, Apr 29 2018, 16:14:56) - [GCC 7.2.0], pyOpenSSL 18.0.0 (OpenSSL 1.0.2o  27 Mar 2018), cryptography 2.2.2, Platform Linux-3.10.0-862.el7.x86_64-x86_64-with-centos-7.5.1804-Core
2020-04-16 19:22:40 [scrapy.crawler] INFO: Overridden settings:
{'BOT_NAME': 'TweetScraper',
 'LOG_LEVEL': 'INFO',
 'NEWSPIDER_MODULE': 'TweetScraper.spiders',
 'SPIDER_MODULES': ['TweetScraper.spiders'],
 'USER_AGENT': 'TweetScraper'}
2020-04-16 19:22:40 [scrapy.extensions.telnet] INFO: Telnet Password: 1fb55da389e595db
2020-04-16 19:22:40 [scrapy.middleware] INFO: Enabled extensions:
['scrapy.extensions.corestats.CoreStats',
 'scrapy.extensions.telnet.TelnetConsole',
 'scrapy.extensions.memusage.MemoryUsage',
 'scrapy.extensions.logstats.LogStats']
2020-04-16 19:22:41 [scrapy.middleware] INFO: Enabled downloader middlewares:
['scrapy.downloadermiddlewares.httpauth.HttpAuthMiddleware',
 'scrapy.downloadermiddlewares.downloadtimeout.DownloadTimeoutMiddleware',
 'scrapy.downloadermiddlewares.defaultheaders.DefaultHeadersMiddleware',
 'scrapy.downloadermiddlewares.useragent.UserAgentMiddleware',
 'scrapy.downloadermiddlewares.retry.RetryMiddleware',
 'scrapy.downloadermiddlewares.redirect.MetaRefreshMiddleware',
 'scrapy.downloadermiddlewares.httpcompression.HttpCompressionMiddleware',
 'scrapy.downloadermiddlewares.redirect.RedirectMiddleware',
 'scrapy.downloadermiddlewares.cookies.CookiesMiddleware',
 'scrapy.downloadermiddlewares.httpproxy.HttpProxyMiddleware',
 'scrapy.downloadermiddlewares.stats.DownloaderStats']
2020-04-16 19:22:41 [scrapy.middleware] INFO: Enabled spider middlewares:
['scrapy.spidermiddlewares.httperror.HttpErrorMiddleware',
 'scrapy.spidermiddlewares.offsite.OffsiteMiddleware',
 'scrapy.spidermiddlewares.referer.RefererMiddleware',
 'scrapy.spidermiddlewares.urllength.UrlLengthMiddleware',
 'scrapy.spidermiddlewares.depth.DepthMiddleware']
Mysql连接成功###################################### MySQLCursorBuffered: (Nothing executed yet)
2020-04-16 19:22:41 [TweetScraper.pipelines] INFO: Table 'tweets' already exists
2020-04-16 19:22:41 [scrapy.middleware] INFO: Enabled item pipelines:
['TweetScraper.pipelines.SavetoMySQLPipeline']
2020-04-16 19:22:41 [scrapy.core.engine] INFO: Spider opened
2020-04-16 19:22:41 [scrapy.extensions.logstats] INFO: Crawled 0 pages (at 0 pages/min), scraped 0 items (at 0 items/min)
2020-04-16 19:22:41 [scrapy.extensions.telnet] INFO: Telnet console listening on 127.0.0.1:6023
2020-04-16 19:23:45 [scrapy.extensions.logstats] INFO: Crawled 1 pages (at 1 pages/min), scraped 11 items (at 11 items/min)
2020-04-16 19:24:44 [scrapy.extensions.logstats] INFO: Crawled 2 pages (at 1 pages/min), scraped 22 items (at 11 items/min)

^C2020-04-16 19:26:27 [scrapy.crawler] INFO: Received SIGINT, shutting down gracefully. Send again to force 
2020-04-16 19:26:27 [scrapy.core.engine] INFO: Closing spider (shutdown)
2020-04-16 19:26:43 [scrapy.extensions.logstats] INFO: Crawled 3 pages (at 1 pages/min), scraped 44 items (at 11 items/min)

file

我们可以看到,该项目运行OK,抓取到的数据也已经被保存在数据库了。

四、清洗数据

因为抓取到的Twitter上有表情等特殊符号,在插入数据库时会报错,所以,这里需要对抓取的内容信息进行清洗。

TweetScraper/utils.py 文件新增filter_emoji过滤方法

import re

def filter_emoji(desstr, restr=''):
    """
    filter emoji
    desstr: origin str
    restr: replace str
    """
    # filter emoji
    try:
        res = re.compile(u'[\U00010000-\U0010ffff]')
    except re.error:
        res = re.compile(u'[\uD800-\uDBFF][\uDC00-\uDFFF]')
    return res.sub(restr, desstr)

TweetCrawler.py 文件中调用该方法:

from TweetScraper.utils import filter_emoji

def parse_tweet_item(self, items):
        for item in items:
            try:
                tweet = Tweet()

                tweet['usernameTweet'] = item.xpath('.//span[@class="username u-dir u-textTruncate"]/b/text()').extract()[0]

                ID = item.xpath('.//@data-tweet-id').extract()
                if not ID:
                    continue
                tweet['ID'] = ID[0]

                ### get text content
                tweet['text'] = ' '.join(
                    item.xpath('.//div[@class="js-tweet-text-container"]/p//text()').extract()).replace(' # ',
                                                                                                        '#').replace(
                    ' @ ', '@')

                ### clear data[20200416]
                # tweet['text'] = re.sub(r"[\s+\.\!\/_,$%^*(+\"\')]+|[+——?【】?~@#¥%……&*]+|\\n+|\\r+|(\\xa0)+|(\\u3000)+|\\t", "", tweet['text']);
                                
                                # 过滤掉表情符号【20200417】
                tweet['text'] = filter_emoji(tweet['text'], '')

                if tweet['text'] == '':
                    # If there is not text, we ignore the tweet
                    continue

                ### get meta data
                tweet['url'] = item.xpath('.//@data-permalink-path').extract()[0]

                nbr_retweet = item.css('span.ProfileTweet-action--retweet > span.ProfileTweet-actionCount').xpath(
                    '@data-tweet-stat-count').extract()
                if nbr_retweet:
                    tweet['nbr_retweet'] = int(nbr_retweet[0])
                else:
                    tweet['nbr_retweet'] = 0

                nbr_favorite = item.css('span.ProfileTweet-action--favorite > span.ProfileTweet-actionCount').xpath(
                    '@data-tweet-stat-count').extract()
                if nbr_favorite:
                    tweet['nbr_favorite'] = int(nbr_favorite[0])
                else:
                    tweet['nbr_favorite'] = 0

                nbr_reply = item.css('span.ProfileTweet-action--reply > span.ProfileTweet-actionCount').xpath(
                    '@data-tweet-stat-count').extract()
                if nbr_reply:
                    tweet['nbr_reply'] = int(nbr_reply[0])
                else:
                    tweet['nbr_reply'] = 0

                tweet['datetime'] = datetime.fromtimestamp(int(
                    item.xpath('.//div[@class="stream-item-header"]/small[@class="time"]/a/span/@data-time').extract()[
                        0])).strftime('%Y-%m-%d %H:%M:%S')

                ### get photo
                has_cards = item.xpath('.//@data-card-type').extract()
                if has_cards and has_cards[0] == 'photo':
                    tweet['has_image'] = True
                    tweet['images'] = item.xpath('.//*/div/@data-image-url').extract()
                elif has_cards:
                    logger.debug('Not handle "data-card-type":\n%s' % item.xpath('.').extract()[0])

                ### get animated_gif
                has_cards = item.xpath('.//@data-card2-type').extract()
                if has_cards:
                    if has_cards[0] == 'animated_gif':
                        tweet['has_video'] = True
                        tweet['videos'] = item.xpath('.//*/source/@video-src').extract()
                    elif has_cards[0] == 'player':
                        tweet['has_media'] = True
                        tweet['medias'] = item.xpath('.//*/div/@data-card-url').extract()
                    elif has_cards[0] == 'summary_large_image':
                        tweet['has_media'] = True
                        tweet['medias'] = item.xpath('.//*/div/@data-card-url').extract()
                    elif has_cards[0] == 'amplify':
                        tweet['has_media'] = True
                        tweet['medias'] = item.xpath('.//*/div/@data-card-url').extract()
                    elif has_cards[0] == 'summary':
                        tweet['has_media'] = True
                        tweet['medias'] = item.xpath('.//*/div/@data-card-url').extract()
                    elif has_cards[0] == '__entity_video':
                        pass  # TODO
                        # tweet['has_media'] = True
                        # tweet['medias'] = item.xpath('.//*/div/@data-src').extract()
                    else:  # there are many other types of card2 !!!!
                        logger.debug('Not handle "data-card2-type":\n%s' % item.xpath('.').extract()[0])

                is_reply = item.xpath('.//div[@class="ReplyingToContextBelowAuthor"]').extract()
                tweet['is_reply'] = is_reply != []

                is_retweet = item.xpath('.//span[@class="js-retweet-text"]').extract()
                tweet['is_retweet'] = is_retweet != []

                tweet['user_id'] = item.xpath('.//@data-user-id').extract()[0]
                yield tweet

                if self.crawl_user:
                    ### get user info
                    user = User()
                    user['ID'] = tweet['user_id']
                    user['name'] = item.xpath('.//@data-name').extract()[0]
                    user['screen_name'] = item.xpath('.//@data-screen-name').extract()[0]
                    user['avatar'] = \
                        item.xpath('.//div[@class="content"]/div[@class="stream-item-header"]/a/img/@src').extract()[0]
                    yield user
            except:
                logger.error("Error tweet:\n%s" % item.xpath('.').extract()[0])
                # raise

通过数据清洗,现在可以正常插入到表里了。

五、翻译成中文

我们可以看到,爬取的数据内容有多个国家的语言,如英文、日语、阿拉伯语、法语等,为了能够知道是什么意思,需要将这些文字翻译成中文,怎么翻译呢?其实很简单,GitHub上有一个开源的Python 谷歌翻译包ssut/py-googletrans ,该项目非常强大,可以自动识别语言并且翻译成我们指定的语言,我们只需安装即可使用。

1、安装

$ pip install googletrans

2、使用

>>> from googletrans import Translator
>>> translator = Translator()
>>> translator.translate('안녕하세요.')
# 
>>> translator.translate('안녕하세요.', dest='ja')
# 
>>> translator.translate('veritas lux mea', src='la')
# 
from googletrans import Translator

destination = 'zh-CN' # 翻译为中文
t = '안녕하세요.'
res = Translator().translate(t, dest=destination).text
 print(res)
你好

3、引用到项目

TweetCrawler.py 文件中调用该方法,并且需要在数据库中新增加一个字段 text_cn

# google translate[20200416]
# @see https://github.com/ssut/py-googletrans
from googletrans import Translator

def parse_tweet_item(self, items):
        for item in items:
            try:
                tweet = Tweet()

                tweet['usernameTweet'] = item.xpath('.//span[@class="username u-dir u-textTruncate"]/b/text()').extract()[0]

                ID = item.xpath('.//@data-tweet-id').extract()
                if not ID:
                    continue
                tweet['ID'] = ID[0]

                ### get text content
                tweet['text'] = ' '.join(
                    item.xpath('.//div[@class="js-tweet-text-container"]/p//text()').extract()).replace(' # ',
                                                                                                        '#').replace(
                    ' @ ', '@')

                ### clear data[20200416]
                # tweet['text'] = re.sub(r"[\s+\.\!\/_,$%^*(+\"\')]+|[+——?【】?~@#¥%……&*]+|\\n+|\\r+|(\\xa0)+|(\\u3000)+|\\t", "", tweet['text']);
                                
                                # 过滤掉表情符号【20200417】
                tweet['text'] = filter_emoji(tweet['text'], '')
                                
                                # 翻译成中文 Translate Chinese【20200417】
                tweet['text_cn'] = Translator().translate(tweet['text'],'zh-CN').text;

                if tweet['text'] == '':
                    # If there is not text, we ignore the tweet
                    continue

                ### get meta data
                tweet['url'] = item.xpath('.//@data-permalink-path').extract()[0]

                nbr_retweet = item.css('span.ProfileTweet-action--retweet > span.ProfileTweet-actionCount').xpath(
                    '@data-tweet-stat-count').extract()
                if nbr_retweet:
                    tweet['nbr_retweet'] = int(nbr_retweet[0])
                else:
                    tweet['nbr_retweet'] = 0

                nbr_favorite = item.css('span.ProfileTweet-action--favorite > span.ProfileTweet-actionCount').xpath(
                    '@data-tweet-stat-count').extract()
                if nbr_favorite:
                    tweet['nbr_favorite'] = int(nbr_favorite[0])
                else:
                    tweet['nbr_favorite'] = 0

                nbr_reply = item.css('span.ProfileTweet-action--reply > span.ProfileTweet-actionCount').xpath(
                    '@data-tweet-stat-count').extract()
                if nbr_reply:
                    tweet['nbr_reply'] = int(nbr_reply[0])
                else:
                    tweet['nbr_reply'] = 0

                tweet['datetime'] = datetime.fromtimestamp(int(
                    item.xpath('.//div[@class="stream-item-header"]/small[@class="time"]/a/span/@data-time').extract()[
                        0])).strftime('%Y-%m-%d %H:%M:%S')

                ### get photo
                has_cards = item.xpath('.//@data-card-type').extract()
                if has_cards and has_cards[0] == 'photo':
                    tweet['has_image'] = True
                    tweet['images'] = item.xpath('.//*/div/@data-image-url').extract()
                elif has_cards:
                    logger.debug('Not handle "data-card-type":\n%s' % item.xpath('.').extract()[0])

                ### get animated_gif
                has_cards = item.xpath('.//@data-card2-type').extract()
                if has_cards:
                    if has_cards[0] == 'animated_gif':
                        tweet['has_video'] = True
                        tweet['videos'] = item.xpath('.//*/source/@video-src').extract()
                    elif has_cards[0] == 'player':
                        tweet['has_media'] = True
                        tweet['medias'] = item.xpath('.//*/div/@data-card-url').extract()
                    elif has_cards[0] == 'summary_large_image':
                        tweet['has_media'] = True
                        tweet['medias'] = item.xpath('.//*/div/@data-card-url').extract()
                    elif has_cards[0] == 'amplify':
                        tweet['has_media'] = True
                        tweet['medias'] = item.xpath('.//*/div/@data-card-url').extract()
                    elif has_cards[0] == 'summary':
                        tweet['has_media'] = True
                        tweet['medias'] = item.xpath('.//*/div/@data-card-url').extract()
                    elif has_cards[0] == '__entity_video':
                        pass  # TODO
                        # tweet['has_media'] = True
                        # tweet['medias'] = item.xpath('.//*/div/@data-src').extract()
                    else:  # there are many other types of card2 !!!!
                        logger.debug('Not handle "data-card2-type":\n%s' % item.xpath('.').extract()[0])

                is_reply = item.xpath('.//div[@class="ReplyingToContextBelowAuthor"]').extract()
                tweet['is_reply'] = is_reply != []

                is_retweet = item.xpath('.//span[@class="js-retweet-text"]').extract()
                tweet['is_retweet'] = is_retweet != []

                tweet['user_id'] = item.xpath('.//@data-user-id').extract()[0]
                yield tweet

                if self.crawl_user:
                    ### get user info
                    user = User()
                    user['ID'] = tweet['user_id']
                    user['name'] = item.xpath('.//@data-name').extract()[0]
                    user['screen_name'] = item.xpath('.//@data-screen-name').extract()[0]
                    user['avatar'] = \
                        item.xpath('.//div[@class="content"]/div[@class="stream-item-header"]/a/img/@src').extract()[0]
                    yield user
            except:
                logger.error("Error tweet:\n%s" % item.xpath('.').extract()[0])
                # raise

items.py 中新增加字段

# -*- coding: utf-8 -*-

# Define here the models for your scraped items
from scrapy import Item, Field

class Tweet(Item):
    ID = Field()       # tweet id
    url = Field()      # tweet url
    datetime = Field() # post time
    text = Field()     # text content
    text_cn = Field()  # text Chinese content  (新增字段)
    user_id = Field()  # user id

管道 piplines.py 文件中修改数据库持久化的方法,新增加text_cn字段

class SavetoMySQLPipeline(object):

    ''' pipeline that save data to mysql '''
    def __init__(self):
        # connect to mysql server
        self.cnx = mysql.connector.connect(
            user=SETTINGS["MYSQL_USER"],
            password=SETTINGS["MYSQL_PWD"],
            host=SETTINGS["MYSQL_SERVER"],
            database=SETTINGS["MYSQL_DB"],
            buffered=True)
        self.cursor = self.cnx.cursor()

        print('Mysql连接成功######################################', self.cursor)
        self.table_name = SETTINGS["MYSQL_TABLE"]
        create_table_query =   "CREATE TABLE `" + self.table_name + "` (\
                `ID` CHAR(20) NOT NULL,\
                `url` VARCHAR(140) NOT NULL,\
                `datetime` VARCHAR(22),\
                `text` VARCHAR(280),\
                `text_cn` VARCHAR(280),\
                `user_id` CHAR(20) NOT NULL,\
                `usernameTweet` VARCHAR(20) NOT NULL\
                )"

        try:
            self.cursor.execute(create_table_query)
        except mysql.connector.Error as err:
            logger.info(err.msg)
        else:
            self.cnx.commit()

    def find_one(self, trait, value):
        select_query =  "SELECT " + trait + " FROM " + self.table_name + " WHERE " + trait + " = " + value + ";"
        try:
            val = self.cursor.execute(select_query)
        except mysql.connector.Error as err:
            return False

        if (val == None):
            return False
        else:
            return True

    def check_vals(self, item):
        ID = item['ID']
        url = item['url']
        datetime = item['datetime']
        text = item['text']
        user_id = item['user_id']
        username = item['usernameTweet']

        if (ID is None):
            return False
        elif (user_id is None):
            return False
        elif (url is None):
            return False
        elif (text is None):
            return False
        elif (username is None):
            return False
        elif (datetime is None):
            return False
        else:
            return True


    def insert_one(self, item):
        ret = self.check_vals(item)

        if not ret:
            return None

        ID = item['ID']
        user_id = item['user_id']
        url = item['url']
        text = item['text']
        text_cn = item['text_cn']

        username = item['usernameTweet']
        datetime = item['datetime']

        insert_query =  'INSERT INTO ' + self.table_name + ' (ID, url, datetime, text, text_cn, user_id, usernameTweet )'
        insert_query += ' VALUES ( %s, %s, %s, %s, %s, %s, %s)'
        insert_query += ' ON DUPLICATE KEY UPDATE'
        insert_query += ' url = %s, datetime = %s, text= %s, text_cn= %s, user_id = %s, usernameTweet = %s'

        try:
            self.cursor.execute(insert_query, (
                ID,
                url,
                datetime,
                text,
                text_cn,
                user_id,
                username,
                url,
                datetime,
                text,
                text_cn,
                user_id,
                username
                ))
        # insert and updadte parameter,so repeat
        except mysql.connector.Error as err:
            logger.info(err.msg)
        else:
            self.cnx.commit()

    def process_item(self, item, spider):
        if isinstance(item, Tweet):
           self.insert_one(dict(item))  # Item is inserted or updated.
           logger.debug("Add tweet:%s" %item['url'])

4、再次运行

然后再次运行该命令:

scrapy crawl TweetScraper -a query="Novel coronavirus,#COVID-19"

file

可以看到数据库中已经将外文翻译成中文了^_^。


相关文章:
Python Scrapy爬虫框架学习
ssut/py-googletrans 谷歌翻译

你可能感兴趣的:(scrapy,python,python爬虫,twitter,翻译)