东方财富股吧标题爬取分析

45个股吧，140万条数据库记录
日期从2018-03-01至2021-03-01共36个月的股吧帖子，
爬取股吧名称、阅读、评论、标题、作者和发帖时间，
并分析总体情绪

亮点回顾

时间问题
获取的时间未加年份，解决方法，观察发现发帖日期月份逐级递减，按获取顺序下一个时间月份在同一年内小于等于上一个月份，设一个变量m储存月份，始值设为12，与获取的最新月份new_m比较，若new_m>m，使当前年份减一；再令m=new_m。
数据去重问题
有时候爬取会因各种问题中断，当你再次续爬时数据会重复，于是我加了一个用于去重的myid
myid = item[‘username’] + str(item[‘mdate’])[3:-4] + title[:100]
思想是，时间地点人物组合，即{谁}在{什么时间}{干了什么}地点没加，但也使每条记录内容保证唯一，大概率去重。
考虑过用每个news的url做主键去重，但是一下url是有重复的
创建的数据表语句如下

create table info_guba
(
    id           int auto_increment
        primary key,
    myid         varchar(255) collate utf8mb4_croatian_ci not null,
    scans        int(8)                                   null,
    comments     int(6)                                   null,
    titles       text collate utf8mb4_croatian_ci         null,
    usernames    varchar(50) collate utf8mb4_croatian_ci  null,
    mdates       int(15)                                  null,
    f_scores     float(12, 10) default 0.0000000000       null,
    polarity     int(1)                                   null,
    company_name varchar(80) collate utf8mb4_croatian_ci  null,
    industry     varchar(80) collate utf8mb4_croatian_ci  null
);

跨年份取月份对应时间戳问题
2018-03至2021-03期间每个月份的情绪指数，要取每个月的一号和下个月的一号时间戳，使能完成区间取值，
我的方法
将38个月份数值储存在一个列表，遍历列表月份，如果月份没到倒数第二位，判断月份是否为12月，是则变量年份减一，拿日期转换时间戳得到时间段较大值，再判断，该月份的下一个月份是不是12月，是则年份减一，日期转换时间戳得到较小值

months = [3, 2, 1, 12, 11, 10, 9, 8, 7, 6, 5,
          4, 3, 2, 1, 12, 11, 10, 9, 8, 7, 6, 
          5, 4, 3, 2, 1, 12, 11, 10, 9, 8, 7, 
          6, 5, 4, 3, 2]
for num, m in enumerate(months):
    if num != 36:
        if m == 12:
        	year -= 1
    	max_timestamp = self.date_to_timestamp(year, m)
    	if months[num + 1] == 12:
        	year02 = year - 1
    	else:
        	year02 = year
    	month = str(year02) + '-' + str(months[num + 1])
    	min_timestamp = self.date_to_timestamp(year02, months[num + 1])
    break

爬虫部分代码

# /usr/bin/python 
# --*-- coding:UTF-8 --*--
# Date 2021/3/18 16:35
import re
import time
import jieba
import pymysql
import requests
import pandas as pd
from lxml import etree
from snownlp import SnowNLP


class guba():
    def __init__(self, host, db, user, passwd):
        self.headers = {'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, '
                                      'like Gecko) Chrome/89.0.4389.90 Safari/537.36 Edg/89.0.774.54'}
        self.host = host
        self.db = db
        self.user = user
        self.passwd = passwd
        self.dataoutput = DataOutput()
        self.ip_num = 1

    # 获取代理
    def get_new_ip(self):
        if self.ip_num <= 1000:
            ip_port = requests.get(
                '获取代理的api',
                timeout=6)
            ip = ip_port.text.replace('\r\n', '')
            proxyip = {"http": "http://" + ip,
                       "https": "https://" + ip}
            self.ip_num += 1
        else:
            return None

        return proxyip

    # 移除换行符
    def rm_special_letters(self, old_list):
        new_list = []
        for i in old_list:
            i = i.replace('\r\n', '')
            i = i.replace(' ', '')
            new_list.append(i)
        return new_list

    # 将日期格式转换为时间戳
    def date_to_timestamp(self, year, timestr):
        mdate = str(year) + '-' + timestr
        time_array = time.strptime(mdate, "%Y-%m-%d")
        news_timestamp = time.mktime(time_array)
        return news_timestamp

    # 获取每日热帖  阅读量 评论量 标题 用户 发帖时间
    def dangu_pinglun(self, url, company_name, industry):
        """
        :param 所属板块:
        :param 公司名称:
        :type url: 股吧首页链接
        """
        global mtimestamp
        mtimestamp = time.time()
        page = 1
        year = 2021
        latest_mounth = 12
        proxyip = self.get_new_ip()
        while True:
            datalist = []
            try:
                if page % 50 == 0:
                    #每50页换一次代理ip 防止反扒
                    print(company_name, page, "----" + str(time.time()))
                    proxyip = self.get_new_ip()
                #拼接url
                murl = url + str(page) + '.html'
                resp = requests.get(murl, headers=self.headers, proxies=proxyip, timeout=10)
                # print(resp.text)
                htmltree = etree.HTML(resp.text)
                yuedu_count = htmltree.xpath('//span[@class="l1 a1"]/text()')
                yuedu_count = self.rm_special_letters(yuedu_count)[1:]
                pinglun_count = htmltree.xpath('//span[@class="l2 a2"]/text()')
                pinglun_count = self.rm_special_letters(pinglun_count)[1:]
                title_list = htmltree.xpath('//span[@class="l3 a3"]/a/@title')
                username_list = htmltree.xpath('//span[@class="l4 a4"]/a//text()')
                last_time_list = htmltree.xpath('//span[@class="l5 a5"]/text()')[1:]

                # 此处将评论列表保存到字典 交给dataoutput储存
                for num, p in enumerate(pinglun_count):
                    # 当阅读数量含有汉字时
                    if re.search('[\u4e00-\u9fa5]', yuedu_count[num]):
                        yuedu_count[num] = 20000
                    if re.search('[\u4e00-\u9fa5]', pinglun_count[num]):
                        pinglun_count[num] = 20000
                    # 截取时间具体提到天 去掉时分时间
                    lastdate = last_time_list[num].split(' ')[0]
                    #发帖时间递减 ，当下层月份大于上边时年份减一
                    if int(lastdate.split('-')[0]) > latest_mounth:
                        year -= 1
                    mtimestamp = self.date_to_timestamp(year, lastdate)
                    info_dict = {'scan': yuedu_count[num],
                                 'comment_num': pinglun_count[num],
                                 'title': title_list[num],
                                 'username': username_list[num],
                                 'mdate': mtimestamp,
                                 'company': company_name,
                                 'industry': industry}
                    datalist.append(info_dict)
                    latest_mounth = int(lastdate.split('-')[0])
                page += 1

                # 将存库语句写到这里是为了个别字节数据储存终端而导致总程序终段
                self.dataoutput.write_to_mysql(host=self.host, db=self.db, user=self.user, passwd=self.passwd,
                                               datalist=datalist)
                time.sleep(1)
            except Exception as e:
                print(industry, company_name, page, "---" + str(time.time()))
                print(str(e))
                if 'HTTPConnectionPool' in str(e):
                    proxyip = self.get_new_ip()
                if 'index out of range' in str(e):
                    page += 1
                elif 'day is out of range for month' in str(e):
                    page += 1
            # 此处判断总时间是否到达最大时间  即到2018年3月终止 爬取下一个
            if mtimestamp <= 1521475200:
                print('时间到')
                break

class DataOutput():
    def __init__(self):
        self.__tablename = 'info_guba'
        self.__tablekeys = '(myid,scans,comments,titles,usernames,mdates,f_scores,company_name,industry)'
	#删除特殊字符 以防引起mysql异常
    def rm_special_letter(self, line):
        for i in ["\'", "\"", "#", "\\"]:
            line = line.replace(i, "")
        return line
        
	"""借助snownlp
    分析news的情绪分为3级 0：积极  1：中立  2：消极"""
    def feeling(self, line):
        try:
            res = SnowNLP(line)
            f_score = res.sentiments
        except:
            f_score = 0
        return f_score

    def __rm_stopwords(self, wordlist):
        new_wordlist = []
        with open('tool_files/stopwords.txt', 'r', encoding='utf-8') as r:
            stopwords = r.read()
            for i in wordlist:
                if i in stopwords:
                    continue
                else:
                    new_wordlist.append(i)
            return new_wordlist
            
	"""使用玻森情感词典 计算情绪指数"""
    def feeling2(self, line):
        path = "tool_files/sentiment_score.txt"
        df = pd.read_table(path, sep=" ", names=['key', 'score_snownlp'])
        key = df['key'].values.tolist()
        score = df['score_snownlp'].values.tolist()

        def getscore(line):
            segs = jieba.lcut(line)  # 分词
            jieba.load_userdict('tool_files/userdict.txt')
            segs = self.__rm_stopwords(segs)
            score_list = [score[key.index(x)] for x in segs if (x in key)]
            # 修改后的sentiment_score.txt 得分有的为字符串格式不能直接使用sum求和
            # print(score_list)
            if len(score_list) != 0:
                sums = 0
                for i in score_list:
                    sums = sums + float(i)
                return sums / len(score_list)  # 计算得分
            else:
                return 0

        last_score = getscore(line)
        if last_score == 0:
            return 0
        else:
            return round(last_score, 5)
	#数据去重
    def __mysql_data_rechecking(self, item, ids_inmysql):
        id_inmysqls = [myid[0] for myid in ids_inmysql]
        title = self.rm_special_letter(item['title'])
        myid = item['username'] + str(item['mdate'])[3:-4] + title[:100]
        if myid not in id_inmysqls:
            return 'newrecord', title, myid
        else:
            return '数据已存在'

    def write_to_mysql(self, datalist, host, db, user, passwd):
        # mysql连接初始化连接
        db = pymysql.connect(host=host, user=user, password=passwd, database=db)
        # 使用 cursor() 方法创建一个游标对象cursor
        cursor = db.cursor()
        # 查询表中 plantform title username 数据拼接字符串用于去重
        quchong_sql = 'SELECT myid FROM {}'.format(self.__tablename)
        cursor.execute(quchong_sql)
        myids = cursor.fetchall()
        for item in datalist:
            data = self.__mysql_data_rechecking(item, myids)
            if data[0] == 'newrecord':
                title, myid = data[1], data[2]
                # feeling = self.feeling(title)
                feeling = 0
                # SQL插入语句
                sql = "INSERT INTO {TABLENAME}{keys}" \
                      "VALUES ('{v0}','{v1}','{v2}','{v3}','{v4}','{v5}','{v6}','{v7}','{v8}')".format \
                    (TABLENAME=self.__tablename,
                     keys=self.__tablekeys,
                     v0=myid,
                     v1=item['scan'],
                     v2=item['comment_num'],
                     v3=title,
                     v4=item['username'],
                     v5=item['mdate'],
                     v6=feeling,
                     v7=item['company'],
                     v8=item['industry'])
                try:
                    # 执行sql语句
                    cursor.execute(sql)
                    # 执行sql语句
                    db.commit()
                except Exception as e:
                    if 'PRIMARY' in str(e):
                        print('查重失败')
                    else:
                        print(item)
                        print(str(e) + "---" + str(time.time()))
                        # 发生错误时回滚
                        db.rollback()
                        # raise e
        # 关闭数据库连接
        db.close()



data01 = {
    '批发和零售业': ['大参林 603233', '广百股份 002187', '来伊份 603777'],
    '制造业': ['中国中车 601766', '永兴材料 002756', '海思科 002653'],
    '房地产业': ['格力地产 600185', '绿景控股 000502', '万科Ａ 000002'],

    '租赁和商务服务业': ['深圳华强 000062', '渤海租赁 000415', '轻纺城 600790'],
    '采矿业': ['兴业矿业 000426', '冀中能源 000937', '中国石化 600028'],
    '交通运输、仓储和邮政业': ['中远海控 601919', '宜昌交运 002627', '大众交通 600611'],
    '信息传输、软件和信息技术服务业': ['恒生电子 600570', '中国联通 600050', '恒华科技 300365'],

    '教育': ['好未来 ustal', '中公教育 002607', '紫光学大 000526'],
    '卫生和社会工作业': ['通策医疗 600763', '迪安诊断 300244', '爱尔眼科 300015'],
    '文化、体育和娱乐业': ['凤凰传媒 601928', '新华传媒 600825', '长江传媒 600757'],
    '金融业': ['民生银行 600016', '中国平安 601318', '国信证券 002736'],

    '建筑业': ['棕榈股份 002431', '上海建工 600170', '隧道股份 600820'],
    '电力、热力、燃气及水的生产和供应业': ['滨海能源 000695', '太阳能 000591', '上海电力 600021'],
    '水利、环境和公共设施管理业': ['远达环保 600292', '碧水源 300070', '启迪环境 000826']
}

if __name__ == '__main__':
    gb = guba(host='localhost', db='guba', user='root', passwd='root')
    for item in data:
        for num, company in enumerate(data[item]):
            stock_code = company.split(' ')[1]
            name = company.split(' ')[0]
            url = 'http://guba.eastmoney.com/list,' + str(stock_code) + ',f_'
            gb.dangu_pinglun(url, name, item)

情绪分析指标计算部分

# /usr/bin/python 
# --*-- coding:UTF-8 --*--
# Date 2021/3/23 21:12
import time
import pymysql
from dataoutput import DataOutput

d = DataOutput()


class Analyse():
    def __init__(self):
        self.host = "localhost"
        self.db = 'guba'
        self.user = 'root'
        self.passwd = 'root'

    # 该部分计算每个title的情绪得分
    def get_score_and_polarity(self):
        db = pymysql.connect(host=self.host, user=self.user, password=self.passwd, database=self.db)
        cursor = db.cursor()
        sql01 = 'select titles,id from info_guba'
        cursor.execute(sql01)
        res = cursor.fetchall()
        for i in res:
            # 使用snownlp计算情绪值
            score_snownlp = d.feeling(i[0])
            # 评出情感极性 将情绪得分>0.6的评论当作积极评论，小于0.4的评论当作消极评论。
            if score_snownlp <= 0.6:
                if score_snownlp > 0.4:
                    p = 0
                else:
                    p = -1
            else:
                p = 1
            sql02 = "update info_guba set f_scores={0},polarity={1} where id={2}".format(score_snownlp, p, i[1])
            cursor.execute(sql02)
            db.commit()
            print(i[1])
        # db.commit()
        db.close()

    # 计算单股吧一天内所有帖子情绪值加和求平均
    def get_score_of_day(self):
        db = pymysql.connect(host=self.host, user=self.user, password=self.passwd, database=self.db)
        cursor = db.cursor()
        sql03 = "select company_name from info_guba group by company_name"
        cursor.execute(sql03)
        com_names = cursor.fetchall()
        for i in com_names:
            # 该语句查询每只股票散户每天发贴的情绪
            sql04 = "select round(sum(f_scores*log10(scans+comments))/count(f_scores),10),mdates from info_guba " \
                    "where company_name='{0}' and usernames not like '%资讯%' group by " \
                    "mdates order by mdates".format(i[0])

            cursor.execute(sql04)
            res = cursor.fetchall()
            for j in res:
                score = j[0]
                day = j[1]
                com_name = i[0]
                sql05 = "insert into score_of_day(score,daytimestamp,company_name) " \
                        "values ({0},{1},'{2}')".format(score, day, com_name)
                # print(sql03)
                cursor.execute(sql05)
                db.commit()
        db.close()

    # 将日期格式转换为时间戳
    def date_to_timestamp(self, year, month):
        mdate = str(year) + '-' + str(month)
        time_array = time.strptime(mdate, "%Y-%m")
        news_timestamp = time.mktime(time_array)
        return news_timestamp

    def month_score_of_all(self):
        db = pymysql.connect(host=self.host, user=self.user, password=self.passwd, database=self.db)
        cursor = db.cursor()
        year = 2021
        months = [4, 3, 2, 1, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 12, 11, 10,
                  9, 8, 7, 6, 5, 4, 3, 2]
        for num, m in enumerate(months):
            if num != 37:
                if m == 12:
                    year -= 1
                max_timestamp = self.date_to_timestamp(year, m)
                if months[num + 1] == 12:
                    year02 = year - 1
                else:
                    year02 = year
                month = str(year02) + '-' + str(months[num + 1])
                min_timestamp = self.date_to_timestamp(year02, months[num + 1])
                sql06 = "select round(sum(polarity*f_scores*log10(scans+comments)),10) from info_guba " \
                        "where usernames not like '%资讯%' " \
                        "and mdates>{0} and mdates<{1}".format \
                    (min_timestamp, max_timestamp)
                cursor.execute(sql06)
                res = cursor.fetchall()
                print(month, res)
                #break
                continue
            break