mlscraper: Scrape data from HTML pages automatically with Machine Learning
)
mlscraper allows you to extract structured data from HTML automatically with Machine Learning. You train it by providing a few examples of your desired output. It will then figure out the extraction rules for you automatically and afterwards you'll be able to extract data from any new page you provide.
Background Story
Many services for crawling and scraping automation allow you to select data in a browser and get JSON results in return. No need to specify CSS selectors or anything else.
I've been wondering for a long time why there's no Open Source solution that does something like this. So here's my attempt at creating a python library to enable automatic scraping.
All you have to do is define some examples of scraped data. mlscraper will figure out everything else and return clean data.
Currently, this is a proof of concept with a simplistic solution.
How it works
After you've defined the data you want to scrape, mlscraper will:
- find your samples inside the HTML DOM
- determine which rules/methods to apply for extraction
- extract the data for you and return it in a dictionary
import requests
from mlscraper import RuleBasedSingleItemScraper
from mlscraper.training import SingleItemPageSample
# the items found on the training page
targets = {
"https://test.com/article/1": {"title": "One great result!", "description": "Some description"},
"https://test.com/article/2": {"title": "Another great result!", "description": "Another description"},
"https://test.com/article/3": {"title": "Result to be found", "description": "Description to crawl"},
}
# fetch html and create samples
samples = [SingleItemPageSample(requests.get(url).content, targets[url]) for url in targets]
# training the scraper with the items
scraper = RuleBasedSingleItemScraper.build(samples)
# apply the learned rules and extract new item automatically
result = scraper.scrape(requests.get('https://test.com/article/4').content)
print(result)
# results in something like:
# {'title': 'Article four', 'description': 'Scraped automatically'}
You can find working scrapers like a stackoverflow and a quotes scraper in the examples folder.
Getting started
Install the library via pip install mlscraper. You can then import it via mlscraper and use it as shown in the examples.
Development
See CONTRIBUTING.rst
Related work
If you're interested in the underlying research, I can highly recommend these publications:
- Learning to extract hierarchical information from semi-structured documents: http://ftp.cse.buffalo.edu/users/azhang/disc/disc01/cd1/out/papers/cikm/p250.pdf
- WHISK: Extraction of structured and unstructured information: https://www.cis.uni-muenchen.de/~yeong/Kurse/ws0809/WebDataMining/whisk.pdf
I originally called this autoscraper but while working on it someone else released a library named exactly the same. Check it out here: autoscraper.
2 Jul 22, 2022
1.5k Dec 30, 2022
11 Nov 15, 2021
12 Oct 27, 2022
1 Dec 29, 2021
5 Nov 29, 2021
1 Nov 07, 2021
6 Dec 06, 2022
4 Feb 13, 2022
2 May 09, 2022
2 Apr 29, 2022
2 Nov 01, 2021
129 Dec 14, 2022
776 Jul 28, 2021
17 Jan 24, 2022
3 Jul 08, 2022
2k May 06, 2021
10 Jul 03, 2022
36 Oct 27, 2022
1 Jan 14, 2022