def extract_date(txt):
# TODO: provide default timezone based on guessed country (prob from domain name)
filler = fuzzydate.fuzzydate( day=1, hour=0, minute=0, second=0, microsecond=0)
fd = fuzzydate.parse_datetime(txt)
try:
return fd.datetime(filler)
except:
return None
def extract_date(txt):
# TODO: provide default timezone based on guessed country (prob from domain name)
filler = fuzzydate.fuzzydate(day=1,
hour=0,
minute=0,
second=0,
microsecond=0)
fd = fuzzydate.parse_datetime(txt)
try:
return fd.datetime(filler)
except:
return None
def testExamplesFromWild(self):
filler = fuzzydate(month=1, day=1, hour=0, minute=0, second=0, microsecond=0)
for foo in self.examples_from_the_wild:
fuzzy = parse_datetime(foo[0])
got = fuzzy.datetime(filler)
if foo[1] is not None:
expected = datetime.datetime(*foo[1])
else:
expected = None
self.assertEqual(got,expected, "'%s': expected '%s', got '%s')" % (foo[0],expected,got))
def extract_pubdate(doc, url, headline_linenum):
""" returns date,linenum """
candidates = {}
logging.debug("extracting pubdate")
# TODO: try some definitive meta tags first?
# "DCSext.articleFirstPublished"
# "DC.date.issued"
# "last-modified"
# check for date in slug
for pat in pats.pubdate['url_datefmts']:
m = pat.search(url)
if m is not None:
d = datetime.datetime( int(m.group('year')), int(m.group('month')), int(m.group('day')) )
logging.debug(" using %s from url" % (d,))
return d,None
meta_dates = set()
for meta in doc.findall('.//meta'):
n = meta.get('name', meta.get('property', ''))
if pats.pubdate['metatags'].search(n):
logging.debug(" date: consider meta name='%s' content='%s'" % (n,meta.get('content','')))
fuzzy = fuzzydate.parse_datetime(meta.get('content',''))
if not fuzzy.empty_date():
meta_dates.add(fuzzy.date(fuzzydate.fuzzydate(day=1)))
# if len(meta_dates)==1:
# # only one likely-looking <meta> entry - lets go with it
# d = list(meta_dates)[0]
# logging.debug(" using %s from <meta>" % (d,))
# return d,None
# start looking through whole page
for e in util.tags(doc,'p','span','div','li','td','th','h4','h5','h6','font'):
txt = unicode(e.text_content()).strip()
txt = u' '.join(txt.split())
# discard anything too short or long
if len(txt)<6 or len(txt) > 150:
continue
score = 1
dt = extract_date(txt)
if dt is None:
continue
logging.debug(" date: considering %s '%s'" % (e.tag,txt))
# TEST: proximity to headline in html
if headline_linenum>0 and e.sourceline>0:
dist = e.sourceline - headline_linenum
if dist >-10 and dist <25:
logging.debug(" near headline")
score += 1
# TEST: likely class or id?
if pats.pubdate['classes'].search(e.get('class','')):
logging.debug(" likely class")
score += 1
if pats.pubdate['classes'].search(e.get('id','')):
logging.debug(" likely id")
score += 1
# in byline is also a good indicator
if pats.byline['classes'].search(e.get('class','')):
logging.debug(" likely class")
score += 1
if pats.byline['classes'].search(e.get('id','')):
logging.debug(" likely id")
score += 1
# TEST: also appears in likely <meta> tags?
if dt.date() in meta_dates:
logging.debug(" appears in <meta>")
score += 1
# TEST: not within likely-looking comment container?
in_comment = False
foo = e.getparent()
while foo is not None:
if pats.pubdate['comment_classes'].search(foo.get('class','')):
in_comment = True
break
foo = foo.getparent()
if not in_comment:
logging.debug(" not inside likely comment")
score += 1
# TEST: indicative text? ("posted on" , "last updated" etc...)
if pats.pubdate['pubdate_indicator'].search(txt):
logging.debug(" text indicative of pubdate")
score += 1
# TEST: date appears in url? eg "http://blah.com/blahblah-20100801-blah.html"
if re.compile("%d[-_/.]?0?%d[-_/.]?0?%d" % (dt.year,dt.month,dt.day)).search(url):
logging.debug(" full date appears in url")
score += 2
elif re.compile("%d[-_/.]?0?%d" % (dt.year,dt.month)).search(url):
logging.debug(" year and month appear in url")
score += 1
if dt.date() not in candidates or score>candidates[dt.date()]['score']:
candidates[dt.date()] = {'datetime': dt, 'score': score, 'node': e}
if not candidates:
return None,None
out = sorted(candidates.items(), key=lambda item: item[1]['score'], reverse=True)
# print "========="
# pprint( out[:5] )
# print "========="
best = out[0][1]
return best['datetime'],best['node']
def extract_pubdate(doc, url, headline_linenum):
""" returns date,linenum """
candidates = {}
logging.debug("extracting pubdate")
# TODO: try some definitive meta tags first?
# "DCSext.articleFirstPublished"
# "DC.date.issued"
# "last-modified"
# check for date in slug
for pat in pats.pubdate['url_datefmts']:
m = pat.search(url)
if m is not None:
d = datetime.datetime(int(m.group('year')), int(m.group('month')),
int(m.group('day')))
logging.debug(" using %s from url" % (d, ))
return d, None
meta_dates = set()
for meta in doc.findall('.//meta'):
n = meta.get('name', meta.get('property', ''))
if pats.pubdate['metatags'].search(n):
logging.debug(" date: consider meta name='%s' content='%s'" %
(n, meta.get('content', '')))
fuzzy = fuzzydate.parse_datetime(meta.get('content', ''))
if not fuzzy.empty_date():
meta_dates.add(fuzzy.date(fuzzydate.fuzzydate(day=1)))
# if len(meta_dates)==1:
# # only one likely-looking <meta> entry - lets go with it
# d = list(meta_dates)[0]
# logging.debug(" using %s from <meta>" % (d,))
# return d,None
# start looking through whole page
for e in util.tags(doc, 'p', 'span', 'div', 'li', 'td', 'th', 'h4', 'h5',
'h6', 'font'):
txt = unicode(e.text_content()).strip()
txt = u' '.join(txt.split())
# discard anything too short or long
if len(txt) < 6 or len(txt) > 150:
continue
score = 1
dt = extract_date(txt)
if dt is None:
continue
logging.debug(" date: considering %s '%s'" % (e.tag, txt))
# TEST: proximity to headline in html
if headline_linenum > 0 and e.sourceline > 0:
dist = e.sourceline - headline_linenum
if dist > -10 and dist < 25:
logging.debug(" near headline")
score += 1
# TEST: likely class or id?
if pats.pubdate['classes'].search(e.get('class', '')):
logging.debug(" likely class")
score += 1
if pats.pubdate['classes'].search(e.get('id', '')):
logging.debug(" likely id")
score += 1
# in byline is also a good indicator
if pats.byline['classes'].search(e.get('class', '')):
logging.debug(" likely class")
score += 1
if pats.byline['classes'].search(e.get('id', '')):
logging.debug(" likely id")
score += 1
# TEST: also appears in likely <meta> tags?
if dt.date() in meta_dates:
logging.debug(" appears in <meta>")
score += 1
# TEST: not within likely-looking comment container?
in_comment = False
foo = e.getparent()
while foo is not None:
if pats.pubdate['comment_classes'].search(foo.get('class', '')):
in_comment = True
break
foo = foo.getparent()
if not in_comment:
logging.debug(" not inside likely comment")
score += 1
# TEST: indicative text? ("posted on" , "last updated" etc...)
if pats.pubdate['pubdate_indicator'].search(txt):
logging.debug(" text indicative of pubdate")
score += 1
# TEST: date appears in url? eg "http://blah.com/blahblah-20100801-blah.html"
if re.compile("%d[-_/.]?0?%d[-_/.]?0?%d" %
(dt.year, dt.month, dt.day)).search(url):
logging.debug(" full date appears in url")
score += 2
elif re.compile("%d[-_/.]?0?%d" % (dt.year, dt.month)).search(url):
logging.debug(" year and month appear in url")
score += 1
if dt.date(
) not in candidates or score > candidates[dt.date()]['score']:
candidates[dt.date()] = {'datetime': dt, 'score': score, 'node': e}
if not candidates:
return None, None
out = sorted(candidates.items(),
key=lambda item: item[1]['score'],
reverse=True)
# print "========="
# pprint( out[:5] )
# print "========="
best = out[0][1]
return best['datetime'], best['node']
