1. 程式人生 > >python爬取亞馬遜簡單的書籍資訊

python爬取亞馬遜簡單的書籍資訊

我有個需求就是抓取一些簡單的書籍資訊儲存到mysql資料庫,例如,封面圖片,書名,型別,作者,簡歷,出版社,語種。

我比較之後,決定在亞馬遜來實現我的需求。

我分析網站後發現,亞馬遜有個高階搜尋的功能,我就通過該搜尋結果來獲取書籍的詳情URL。

由於亞馬遜的高階搜尋是用get方法的,所以通過分析,搜尋結果的URL,可得到node引數是代表書籍型別的。field-binding_browse-bin是代表書籍裝飾。

所以我固定了書籍裝飾為平裝,而書籍的型別,只能每次執行的時候,爬取一種型別的書籍難過

之後就是根據書籍詳情頁面利用正則表示式來匹配需要的資訊了。

以下原始碼,命名不是很規範。。。

import requests
import sys
import re
import pymysql

class product:
    type="歷史"
    name=""
    author=""
    desciption=""
    pic1=""
    languages=""
    press=""

def getProUrl():
    urlList = []
    headers = {"User-Agent":"Mozilla/5.0 (Windows NT 10.0; WOW64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/50.0.2661.102 Safari/537.36"
} session = requests.Session() furl="https://www.amazon.cn/gp/search/ref=sr_adv_b/?search-alias=stripbooks&field-binding_browse-bin=2038564051&sort=relevancerank&page=" for i in range(1,7): html="" print(furl+str(i)) html = session.post(furl+str(i)+'&node=658418051'
,headers = headers) html.encoding = 'utf-8' s=html.text.encode('gb2312','ignore').decode('gb2312') url=r'</li><li id=".*?" data-asin="(.+?)" class="s-result-item celwidget">' reg=re.compile(url,re.M) items = reg.findall(html.text) for i in range(0,len(items)): urlList.append(items[i]) urlList=set(urlList) return urlList def getProData(url): pro = product() headers = {"User-Agent":"Mozilla/5.0 (Windows NT 10.0; WOW64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/50.0.2661.102 Safari/537.36"} session = requests.Session() zurl="https://www.amazon.cn/dp/" html = session.get(zurl+url,headers = headers) html.encoding = 'utf-8' s=html.text.encode('gb2312','ignore').decode('gb2312') pro.pic1=getProPic(html) pro.name=getProName(html) pro.author=getProAuthor(html) pro.desciption=getProDescrip(html) pro.press=getProPress(html) pro.languages=getProLanguages(html) return pro def getProPic(html): pic=r'id="imgBlkFront" data-a-dynamic-image="{&quot;(.+?)&quot;.*?}"' reg=re.compile(pic,re.M) items = reg.findall(html.text) if len(items)==0: return "" else: return items[0] def getProName(html): name=r'<div class="ma-title"><p class="wraptext goto-top">(.+?)<span' reg=re.compile(name,re.M) items = reg.findall(html.text) if len(items)==0: return "" else: return items[0] def getProAuthor(html): author=r'<span class="author.{0,20}" data-width="".{0,30}>.*?<a class="a-link-normal" href=".*?books">(.+?)</a>.*?<span class="a-color-secondary">(.+?)</span>' reg=re.compile(author,re.S) items = reg.findall(html.text) au="" for i in range(0,len(items)): au=au+items[i][0]+items[i][1] return au def getProDescrip(html): Descrip=r'<noscript>.{0,30}<div>(.+?)</div>.{0,30}<em></em>.{0,30}</noscript>.{0,30}<div id="outer_postBodyPS"' reg=re.compile(Descrip,re.S) items = reg.findall(html.text) if len(items)==0: return "" else: position = items[0].find('海報:') descrip=items[0] if position != -1: descrip=items[0][0:position] return descrip.strip() def getProPress(html): press=r'<li><b>出版社:</b>(.+?)</li>' reg=re.compile(press,re.M) items = reg.findall(html.text) if len(items)==0: return "" else: return items[0].strip() def getProLanguages(html): languages=r'<li><b>語種:</b>(.+?)</li>' reg=re.compile(languages,re.M) items = reg.findall(html.text) if len(items)==0: return "" else: return items[0].strip() def getConnection(): config = { 'host':'121.**.**.**', 'port':3306, 'user':'root', 'password':'******', 'db':'home_work', 'charset':'utf8', 'cursorclass':pymysql.cursors.DictCursor, } connection = pymysql.connect(**config) return connection urlList = getProUrl() i = 0 for d in urlList: i = i + 1 print (i) connection = getConnection() pro = getProData(d) try: with connection.cursor() as cursor: sql='INSERT INTO books (type,name,author,desciption,pic1,languages,press) VALUES (%s,%s,%s,%s,%s,%s,%s)' cursor.execute(sql,(pro.type,pro.name,pro.author,pro.desciption,pro.pic1,pro.languages,pro.press)) connection.commit() finally: connection.close();