Skip to content

🤖 Scrape data from HTML websites automatically by just providing examples

Notifications You must be signed in to change notification settings

lorey/mlscraper

Repository files navigation

mlscraper: Scrape data from HTML pages automatically

CI status PyPI version PyPI python version

mlscraper allows you to extract structured data from HTML automatically instead of manually specifying nodes or css selectors. 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.

Image showing how mlscraper turns html into data objects

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.

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

Getting started

mlscraper is currently shortly before version 1.0. If you want to check the new release, use pip install --pre mlscraper to test the release candidate. You can also install the latest (unstable) development version of mlscraper via pip install git+https://github.com/lorey/mlscraper#egg=mlscraper, e.g. to check new features or to see if a bug has been fixed already. Please note that until the 1.0 release pip install mlscraper will return an outdated 0.* version.

To get started with a simple scraped, check out a basic sample below.

import requests
from mlscraper.html import Page
from mlscraper.samples import Sample, TrainingSet
from mlscraper.training import train_scraper

# fetch the page to train
einstein_url = 'http://quotes.toscrape.com/author/Albert-Einstein/'
resp = requests.get(einstein_url)
assert resp.status_code == 200

# create a sample for Albert Einstein
# please add at least two samples in practice to get meaningful rules!
training_set = TrainingSet()
page = Page(resp.content)
sample = Sample(page, {'name': 'Albert Einstein', 'born': 'March 14, 1879'})
training_set.add_sample(sample)

# train the scraper with the created training set
scraper = train_scraper(training_set)

# scrape another page
resp = requests.get('http://quotes.toscrape.com/author/J-K-Rowling')
result = scraper.get(Page(resp.content))
print(result)
# returns {'name': 'J.K. Rowling', 'born': 'July 31, 1965'}

Check the examples directory for usage examples until further documentation arrives.

Development

See CONTRIBUTING.rst

Related work

I originally called this autoscraper but while working on it someone else released a library named exactly the same. Check it out here: autoscraper. Also, while initially driven by Machine Learning, using statistics to search for heuristics turned out to be faster and requires less training data. But since the name is memorable, I'll keep it.