Quickstart

The date guesser uses both the url and the html to work, and uses some heuristics to decide which of many possible dates might be the best one.

from date_guesser import DateGuesser, Accuracy

guesser = DateGuesser()

# Uses url slugs when available
guess = guesser.guess_date(url='https://www.nytimes.com/2017/10/13/some_news.html',
                           html='<could be anything></could>')

#  Returns a namedtuple with three fields
guess.date      # datetime.datetime(2017, 10, 13, 0, 0, tzinfo=<UTC>)
guess.accuracy  # Accuracy.DATE
guess.method    # 'Found /2017/10/13/ in url'

In case there are two trustworthy sources of dates, date_guesser prefers the more accurate one

html = '''
    <html><head>
    <meta property="article:published" itemprop="datePublished" content="2017-10-13T04:56:54-04:00" />
    </head></html>'''
guess = guesser.guess_date(url='https://www.nytimes.com/2017/10/some_news.html',
                           html=html)
guess.date  # datetime.datetime(2017, 10, 13, 4, 56, 54, tzinfo=tzoffset(None, -14400))
guess.accuracy is Accuracy.DATETIME  # True

But date_guesser is not led astray by more accurate, less trustworthy sources of information

html = '''
    <html><head>
    <meta property="og:image" content="foo.com/2016/7/4/whatever.jpg"/>
    </head></html>'''
guess = guesser.guess_date(url='https://www.nytimes.com/2017/10/some_news.html',
                           html=html)
guess.date  # datetime.datetime(2017, 10, 15, 0, 0, tzinfo=<UTC>)
guess.accuracy is Accuracy.PARTIAL  # True

Installation

The library is not yet available on PyPI, so installation is via github only for now:

pip install git+https://github.com/mitmedialab/date_guesser

Performance

We benchmarked the accuracy against the wonderful newspaper library, using one hundred urls gathered from each of four very different topics in the mediacloud system. This includes blogs and news articles, as well as many urls that have no date (in which case a guess is marked correct only if it returns None).