有没有办法优化for循环？ Selenium 抓取 38 页的时间太长

Question

我正在尝试通过 Selenium 和 python 抓取 https://arxiv.org/search/?query=healthcare&searchtype=allI。 for 循环执行时间过长。我尝试使用无头浏览器和 PhantomJS 进行抓取，但它没有抓取抽象字段（需要通过单击更多按钮来扩展抽象字段）

import pandas as pd
import selenium
import re
import time
from selenium.common.exceptions import NoSuchElementException
from selenium.webdriver import Firefox

browser = Firefox()
url_healthcare = 'https://arxiv.org/search/?query=healthcare&searchtype=all'
browser.get(url_healthcare)

dfs = []
for i in range(1, 39):
    articles = browser.find_elements_by_tag_name('li[class="arxiv-result"]')

    for article in articles:
        title = article.find_element_by_tag_name('p[class="title is-5 mathjax"]').text
        arxiv_id = article.find_element_by_tag_name('a').text.replace('arXiv:','')
        arxiv_link = article.find_elements_by_tag_name('a')[0].get_attribute('href') 
        pdf_link = article.find_elements_by_tag_name('a')[1].get_attribute('href')
        authors = article.find_element_by_tag_name('p[class="authors"]').text.replace('Authors:','')

        try:
                link1 = browser.find_element_by_link_text('▽ More')
                link1.click()
        except:
                time.sleep(0.1)

        abstract = article.find_element_by_tag_name('p[class="abstract mathjax"]').text
        date = article.find_element_by_tag_name('p[class="is-size-7"]').text
        date = re.split(r"Submitted|;",date)[1]
        tag = article.find_element_by_tag_name('div[class="tags is-inline-block"]').text.replace('\n', ',')
        
        try:
            doi = article.find_element_by_tag_name('div[class="tags has-addons"]').text
            doi = re.split(r'\s', doi)[1] 
        except NoSuchElementException:
            doi = 'None'

        all_combined = [title, arxiv_id, arxiv_link, pdf_link, authors, abstract, date, tag, doi]
        dfs.append(all_combined)

    print('Finished Extracting Page:', i)

    try:
        link2 = browser.find_element_by_class_name('pagination-next')
        link2.click()
    except:
        browser.close
        
    time.sleep(0.1)

Answer 1

以下实现在 16 秒内实现了这一目标。

为了加快执行过程，我采取了以下措施：

• 完全删除了Selenium（无需点击）
• 对于abstract，使用BeautifulSoup 的输出并稍后处理
• 添加了multiprocessing 以显着加快进程

from multiprocessing import Process, Manager
import requests 
from bs4 import BeautifulSoup
import re
import time

start_time = time.time()

def get_no_of_pages(showing_text):
    no_of_results = int((re.findall(r"(\d+,*\d+) results for all",showing_text)[0].replace(',','')))
    pages = no_of_results//200 + 1
    print("total pages:",pages)
    return pages 

def clean(text):
    return text.replace("\n", '').replace("  ",'')

def get_data_from_page(url,page_number,data):
    print("getting page",page_number)
    response = requests.get(url+"start="+str(page_number*200))
    soup = BeautifulSoup(response.content, "lxml")
    
    arxiv_results = soup.find_all("li",{"class","arxiv-result"})

    for arxiv_result in arxiv_results:
        paper = {} 
        paper["titles"]= clean(arxiv_result.find("p",{"class","title is-5 mathjax"}).text)
        links = arxiv_result.find_all("a")
        paper["arxiv_ids"]= links[0].text.replace('arXiv:','')
        paper["arxiv_links"]= links[0].get('href')
        paper["pdf_link"]= links[1].get('href')
        paper["authors"]= clean(arxiv_result.find("p",{"class","authors"}).text.replace('Authors:',''))

        split_abstract = arxiv_result.find("p",{"class":"abstract mathjax"}).text.split("▽ More\n\n\n",1)
        if len(split_abstract) == 2:
            paper["abstract"] = clean(split_abstract[1].replace("△ Less",''))
        else: 
            paper["abstract"] = clean(split_abstract[0].replace("△ Less",''))

        paper["date"] = re.split(r"Submitted|;",arxiv_results[0].find("p",{"class":"is-size-7"}).text)[1]
        paper["tag"] = clean(arxiv_results[0].find("div",{"class":"tags is-inline-block"}).text) 
        doi = arxiv_results[0].find("div",{"class":"tags has-addons"})       
        if doi is None:
            paper["doi"] = "None"
        else:
            paper["doi"] = re.split(r'\s', doi.text)[1] 

        data.append(paper)
    
    print(f"page {page_number} done")


if __name__ == "__main__":
    url = 'https://arxiv.org/search/?searchtype=all&query=healthcare&abstracts=show&size=200&order=-announced_date_first&'

    response = requests.get(url+"start=0")
    soup = BeautifulSoup(response.content, "lxml")

    with Manager() as manager:
        data = manager.list()  
        processes = []
        get_data_from_page(url,0,data)


        showing_text = soup.find("h1",{"class":"title is-clearfix"}).text
        for i in range(1,get_no_of_pages(showing_text)):
            p = Process(target=get_data_from_page, args=(url,i,data))
            p.start()
            processes.append(p)

        for p in processes:
            p.join()

        print("Number of entires scraped:",len(data))

        stop_time = time.time()

        print("Time taken:", stop_time-start_time,"seconds")

输出：

>>> python test.py

getting page 0
page 0 done
total pages: 10
getting page 1
getting page 4
getting page 2
getting page 6
getting page 5
getting page 3
getting page 7
getting page 9
getting page 8
page 9 done
page 4 done
page 1 done
page 6 done
page 2 done
page 7 done
page 3 done
page 5 done
page 8 done
Number of entires scraped: 1890
Time taken: 15.911492586135864 seconds

注意：

• 请将以上代码写入.py 文件。对于 Jupyter 笔记本，请参阅 this。
• 多处理代码取自here。
• data 列表中条目的顺序与网站上的顺序不匹配，因为 Manager 将在它们出现时将 dictionaries 添加到其中。
• 上述代码自行查找页数，因此可以推广到任何 arxiv 搜索结果。不幸的是，首先要做到这一点getspage 0，然后计算number of pages，然后为剩余页面计算multiprocessing。这样做的缺点是在处理 0th page 时，没有其他进程正在运行。因此，如果您删除该部分并简单地为 10 pages 运行循环，那么所用时间应该会下降到大约 8 秒。

Answer 2

您可以尝试使用请求和美丽的汤方法。无需点击更多链接。

from requests import get
from bs4 import BeautifulSoup

# you can change the size to retrieve all the results at one shot.

url = 'https://arxiv.org/search/?query=healthcare&searchtype=all&abstracts=show&order=-announced_date_first&size=50&start=0'
response = get(url,verify = False)
soup = BeautifulSoup(response.content, "lxml")
#print(soup)
queryresults = soup.find_all("li", attrs={"class": "arxiv-result"})

for result in queryresults:
    title = result.find("p",attrs={"class": "title is-5 mathjax"})
    print(title.text)

#If you need full abstract content - try this (you do not need to click on more button
    for result in queryresults:
        abstractFullContent = result.find("span",attrs={"class": "abstract-full has-text-grey-dark mathjax"})
        print(abstractFullContent.text)

输出：

 Interpretable Deep Learning for Automatic Diagnosis of 12-lead Electrocardiogram
            
  Leveraging Technology for Healthcare and Retaining Access to Personal Health Data to Enhance Personal Health and Well-being
  Towards new forms of particle sensing and manipulation and 3D imaging on a smartphone for healthcare applications