How to reduce amount of scraping time when using Requests-HTML? - selenium

Question

I currently use Requests-HTML version 0.10.0 and Selenium 3.141.0. My project is to scrape the ratings of all articles on this website https://openreview.net/group?id=ICLR.cc/2021/Conference. To open each page of the website (the website has 53 pages and each page has 50 articles), I use Selenium. Next, to open articles on each page, I use Requests-HTML. My question is about how to reduce the time uses to open each article and get the rating. In this case, I use await r_inside.html.arender(sleep = 5, timeout=100), which means the sleeping time is 5 seconds and the timeout is 100 seconds. When I try to reduce sleep time to 0.5 seconds, it will cause an error, which is because it does not have enough time to scrape the website. However, if I keep the sleep time as 5 seconds, it will take 6 to 13 hours to scrape all 2600 articles. Also, after waiting for 13 hours, I can scrape all 2600 articles, but the codes use 88 GB of RAM, which I do not prefer because I need to send this code to other people who will not have enough RAM to run. My purpose is to reduce the scraping time and RAM memory. Below is the code I use.
import csv
link = 'https://openreview.net/group?id=ICLR.cc/2021/Conference'
from requests_html import HTMLSession, AsyncHTMLSession
import time
from tqdm import tqdm
from selenium import webdriver
from selenium.webdriver.common.by import By
from selenium.webdriver.support import expected_conditions as EC
from selenium.webdriver.support.wait import WebDriverWait
id_list = []
keyword_list = []
abstract_list = []
title_list = []
driver = webdriver.Chrome('./requests_html/chromedriver.exe')
driver.get('https://openreview.net/group?id=ICLR.cc/2021/Conference')
cond = EC.presence_of_element_located((By.XPATH, '//*[#id="all-submissions"]/nav/ul/li[13]/a'))
WebDriverWait(driver, 10).until(cond)
for page in tqdm(range(1, 54)):
text = ''
elems = driver.find_elements_by_xpath('//*[#id="all-submissions"]/ul/li')
for i, elem in enumerate(elems):
try:
# parse title
title = elem.find_element_by_xpath('./h4/a[1]')
link = title.get_attribute('href')
paper_id = link.split('=')[-1]
title = title.text.strip().replace('\t', ' ').replace('\n', ' ')
# show details
elem.find_element_by_xpath('./a').click()
time.sleep(0.2)
# parse keywords & abstract
items = elem.find_elements_by_xpath('.//li')
keyword = ''.join([x.text for x in items if 'Keywords' in x.text])
abstract = ''.join([x.text for x in items if 'Abstract' in x.text])
keyword = keyword.strip().replace('\t', ' ').replace('\n', ' ').replace('Keywords: ', '')
abstract = abstract.strip().replace('\t', ' ').replace('\n', ' ').replace('Abstract: ', '')
text += paper_id+'\t'+title+'\t'+link+'\t'+keyword+'\t'+abstract+'\n'
title_list.append(title)
id_list.append(paper_id)
keyword_list.append(keyword)
abstract_list.append(abstract)
except Exception as e:
print(f'page {page}, # {i}:', e)
continue
# next page
try:
driver.find_element_by_xpath('//*[#id="all-submissions"]/nav/ul/li[13]/a').click()
time.sleep(2) # NOTE: increase sleep time if needed
except:
print('no next page, exit.')
break
csv_file = open('./requests_html/bb_website_scrap.csv','w', encoding="utf-8")
csv_writer = csv.writer(csv_file)
csv_writer.writerow(['Title','Keyword','Abstract','Link','Total Number of Reviews','Average Rating','Average Confidence'])
n = 0
for item in range(len(id_list)):
title = title_list[item]
keyword = keyword_list[item]
abstract = abstract_list[item]
id = id_list[item]
link_pdf = f'https://openreview.net/forum?id={id}'
print(id)
asession_inside = AsyncHTMLSession()
r_inside = await asession_inside.get(link_pdf)
print(type(r_inside))
await r_inside.html.arender(sleep = 5, timeout=100)
test_rating = r_inside.html.find('div.comment-level-odd div.note_contents span.note_content_value')
print(len(test_rating))
check_list = {'0', '1', '2', '3', '4', '5', '6', '7', '8', '9','10'}
total_rating_confidence = []
total_rating = []
total_confidence = []
for t in range(len(test_rating)):
if any(test_rating[t].text.split(':')[0] in s for s in check_list):
total_rating_confidence.append(test_rating[t].text.split(':')[0])
for r in range(len(total_rating_confidence)):
if (r % 2 == 0):
total_rating.append(int(total_rating_confidence[r]))
else:
total_confidence.append(int(total_rating_confidence[r]))
average_rating = sum(total_rating) / len(total_rating)
average_confidence = sum(total_confidence) / len(total_confidence)
csv_writer.writerow([title, keyword, abstract, link_pdf,len(total_rating),average_rating,average_confidence])
n = n + 1
print('Order {}',n)
csv_file.close()

Answer 1

I'm no Python expert (in fact, I'm a rank beginner) but the simple answer is better parallelism & session management.
The useful answer is a bit more complicated.
You're leaving the Chromium session around, which is likely what's hoovering up all your RAM. If you call asession_inside.close(), you may see an improvement in RAM usage.
As far as I can tell, you're doing everything in serial; You fetch each page and extract data on the articles in serial. Then, you query each article in serial as well.
You're using arender to fetch each article asynchronously, but you're awaiting it & using a standard for loop. As far as I understand, that means you're not getting any advantage from async; You're still processing each page one at a time (which explains your long process time).
I'd suggest using asyncio to turn the for loop into a parallel version of itself as suggested in this article. Make sure you set a task limit so that you don't try to load all the articles at once; That will also help with your RAM usage.