1

I have a dataset of headlines, such as

http://www.stackoverflow.com/lifestyle/tech/this-is-a-very-nice-headline-my-friend/2013/04/26/acjhrjk-2e1-1krjke4-9el8c-2eheje_story.html?tid=sm_fb

http://www.stackoverflow.com/2015/07/15/sports/baseball/another-very-nice.html?smid=tw-somedia&seid=auto

http://worldnews.stack.com/news/2013/07/22/54216-hello-another-one-here?lite

http://www.stack.com/article_email/hello-one-here-that-is-cool-1545545554-lMyQjAxMTAHFJELMDgxWj

http://www.stack.com/2013/11/13/tech/tricky-one/the-real-one/index.html

http://www.stack.com/2013/11/13/tech/the-good-one.html

http://www.stack.com/news/science-and-technology/54512-hello-world-here-is-a-weird-character#b02g07f20b14

I need to extract from these kind of links the proper headline, that is:

  • this-is-a-very-nice-headline-my-friend
  • another-very-nice
  • hello-another-one-here
  • hello-one-here-that-is-cool
  • the-real-one
  • the-good-one
  • hello-world-here-is-a-weird-character

so the rule seems to find the longest string of the form word1-word2-word3- that has a / at the right or left border and without considering

  1. words with more than 3 digits (for instance acjhrjk-2e1-1krjke4-9el8c-2eheje in the first link, or 54216 in the third one ,
  2. excluding stuff like .html.

How can I do that using regex in Python? I believe regex is the only viable solution here unfortunately. Packages such as yurl or urlparse can capture the path of the url, but then I am back to using regex to get the headline..

Many thanks!

ℕʘʘḆḽḘ
  • 18,566
  • 34
  • 128
  • 235

1 Answers1

1

After all, regular expressions might not be your best bet.
However, with the specifications you came up with, you could do the following:

import re

urls = ['http://www.stackoverflow.com/lifestyle/tech/this-is-a-very-nice-headline-my-friend/2013/04/26/acjhrjk-2e1-1krjke4-9el8c-2eheje_story.html?tid=sm_fb',
'http://www.stackoverflow.com/2015/07/15/sports/baseball/another-very-nice.html?smid=tw-somedia&seid=auto',
'http://worldnews.stack.com/news/2013/07/22/54216-hello-another-one-here?lite',
'http://www.stack.com/article_email/hello-one-here-that-is-cool-1545545554-lMyQjAxMTAHFJELMDgxWj',
'http://www.stack.com/2013/11/13/tech/tricky-one/the-real-one/index.html',
'http://www.stack.com/2013/11/13/tech/the-good-one.html',
'http://www.stack.com/news/science-and-technology/54512-hello-world-here-is-a-weird-character#b02g07f20b14']

regex = re.compile(r'(?<=/)([-\w]+)(?=[.?/#]|$)')
digits = re.compile(r'-?\d{3,}-?')

for url in urls:
    substrings = regex.findall(url)
    longest = max(substrings, key=len)
    headline = re.sub(digits, '', longest)
    print headline

This will print 
 this-is-a-very-nice-headline-my-friend
 another-very-nice
 hello-another-one-here
 hello-one-here-that-is-coollMyQjAxMTAHFJELMDgxWj
 the-real-one
 the-good-one
 hello-world-here-is-a-weird-character

See a demo on ideone.com.

Explanation

Here, the regex uses lookarounds to look for a / behind and one of .?/# ahead. Any word character and dash in between is captured.
This is not very specific but if you're looking for the longest substring and eliminate more then three consecutive digits afterwards, it might be a good starting point.
As already said in the comments, you might perhaps be better off using linguistic tools.

Jan
  • 42,290
  • 8
  • 54
  • 79