#!/usr/bin/env python
from cleaners import html_cleaner, clean_attributes
from collections import defaultdict
from htmls import build_doc, get_body, get_title, shorten_title
from lxml.etree import tostring, tounicode
from lxml.html import fragment_fromstring, document_fromstring
import logging
import re
import sys
logging.basicConfig(level=logging.INFO)
REGEXES = {
'unlikelyCandidatesRe': re.compile('combx|comment|community|disqus|extra|foot|header|menu|remark|rss|shoutbox|sidebar|sponsor|ad-break|agegate|pagination|pager|popup|tweet|twitter',re.I),
'okMaybeItsACandidateRe': re.compile('and|article|body|column|main|shadow',re.I),
'positiveRe': re.compile('article|body|content|entry|hentry|main|page|pagination|post|text|blog|story',re.I),
'negativeRe': re.compile('combx|comment|com-|contact|foot|footer|footnote|masthead|media|meta|outbrain|promo|related|scroll|shoutbox|sidebar|sponsor|shopping|tags|tool|widget',re.I),
'divToPElementsRe': re.compile('<(a|blockquote|dl|div|img|ol|p|pre|table|ul)',re.I),
#'replaceBrsRe': re.compile('(]*>[ \n\r\t]*){2,}',re.I),
#'replaceFontsRe': re.compile('<(\/?)font[^>]*>',re.I),
#'trimRe': re.compile('^\s+|\s+$/'),
#'normalizeRe': re.compile('\s{2,}/'),
#'killBreaksRe': re.compile('((\s| ?)*){1,}/'),
#'videoRe': re.compile('http:\/\/(www\.)?(youtube|vimeo)\.com', re.I),
#skipFootnoteLink: /^\s*(\[?[a-z0-9]{1,2}\]?|^|edit|citation needed)\s*$/i,
}
def describe(node, depth=1):
if not hasattr(node, 'tag'):
return "[%s]" % type(node)
name = node.tag
if node.get('id', ''): name += '#'+node.get('id')
if node.get('class', ''):
name += '.' + node.get('class').replace(' ','.')
if name[:4] in ['div#', 'div.']:
name = name[3:]
if depth and node.getparent() is not None:
return name+' - '+describe(node.getparent(), depth-1)
return name
def to_int(x):
if not x: return None
x = x.strip()
if x.endswith('px'):
return int(x[:-2])
if x.endswith('em'):
return int(x[:-2]) * 12
return int(x)
def clean(text):
text = re.sub('\s*\n\s*', '\n', text)
text = re.sub('[ \t]{2,}', ' ', text)
return text.strip()
def text_length(i):
return len(clean(i.text_content() or ""))
class Unparseable(ValueError):
pass
class Document:
TEXT_LENGTH_THRESHOLD = 25
RETRY_LENGTH = 250
def __init__(self, input, **options):
self.input = input
self.options = defaultdict(lambda: None)
for k, v in options.items():
self.options[k] = v
self.html = None
def _html(self, force=False):
if force or self.html is None:
self.html = self._parse(self.input)
return self.html
def _parse(self, input):
doc = build_doc(input)
doc = html_cleaner.clean_html(doc)
base_href = self.options['url']
if base_href:
doc.make_links_absolute(base_href, resolve_base_href=True)
else:
doc.resolve_base_href()
return doc
def content(self):
return get_body(self._html(True))
def title(self):
return get_title(self._html(True))
def short_title(self):
return shorten_title(self._html(True))
def summary(self):
try:
ruthless = True
while True:
self._html(True)
for i in self.tags(self.html, 'script', 'style'):
i.drop_tree()
for i in self.tags(self.html, 'body'):
i.set('id', 'readabilityBody')
if ruthless:
self.remove_unlikely_candidates()
self.transform_misused_divs_into_paragraphs()
candidates = self.score_paragraphs()
best_candidate = self.select_best_candidate(candidates)
if best_candidate:
article = self.get_article(candidates, best_candidate)
else:
if ruthless:
logging.debug("ruthless removal did not work. ")
ruthless = False
self.debug("ended up stripping too much - going for a safer _parse")
# try again
continue
else:
logging.debug("Ruthless and lenient parsing did not work. Returning raw html")
article = self.html.find('body')
if article is None:
article = self.html
cleaned_article = self.sanitize(article, candidates)
of_acceptable_length = len(cleaned_article or '') >= (self.options['retry_length'] or self.RETRY_LENGTH)
if ruthless and not of_acceptable_length:
ruthless = False
continue # try again
else:
return cleaned_article
except StandardError, e:
#logging.exception('error getting summary: ' + str(traceback.format_exception(*sys.exc_info())))
logging.exception('error getting summary: ' )
raise Unparseable(str(e)), None, sys.exc_info()[2]
def get_article(self, candidates, best_candidate):
# Now that we have the top candidate, look through its siblings for content that might also be related.
# Things like preambles, content split by ads that we removed, etc.
sibling_score_threshold = max([10, best_candidate['content_score'] * 0.2])
output = document_fromstring('')
best_elem = best_candidate['elem']
for sibling in best_elem.getparent().getchildren():
#if isinstance(sibling, NavigableString): continue#in lxml there no concept of simple text
append = False
if sibling is best_elem:
append = True
sibling_key = sibling #HashableElement(sibling)
if sibling_key in candidates and candidates[sibling_key]['content_score'] >= sibling_score_threshold:
append = True
if sibling.tag == "p":
link_density = self.get_link_density(sibling)
node_content = sibling.text or ""
node_length = len(node_content)
if node_length > 80 and link_density < 0.25:
append = True
elif node_length < 80 and link_density == 0 and re.search('\.( |$)', node_content):
append = True
if append:
output.append(sibling)
#if output is not None:
# output.append(best_elem)
return output
def select_best_candidate(self, candidates):
sorted_candidates = sorted(candidates.values(), key=lambda x: x['content_score'], reverse=True)
for candidate in sorted_candidates[:5]:
elem = candidate['elem']
self.debug("Top 5 : %6.3f %s" % (candidate['content_score'], describe(elem)))
if len(sorted_candidates) == 0:
return None
best_candidate = sorted_candidates[0]
return best_candidate
def get_link_density(self, elem):
link_length = 0
for i in elem.findall(".//a"):
link_length += text_length(i)
#if len(elem.findall(".//div") or elem.findall(".//p")):
# link_length = link_length
total_length = text_length(elem)
return float(link_length) / max(total_length, 1)
def score_paragraphs(self, ):
MIN_LEN = self.options.get('min_text_length', self.TEXT_LENGTH_THRESHOLD)
candidates = {}
#self.debug(str([describe(node) for node in self.tags(self.html, "div")]))
ordered = []
for elem in self.tags(self.html, "p", "pre", "td"):
parent_node = elem.getparent()
if parent_node is None:
continue
grand_parent_node = parent_node.getparent()
inner_text = clean(elem.text_content() or "")
inner_text_len = len(inner_text)
# If this paragraph is less than 25 characters, don't even count it.
if inner_text_len < MIN_LEN:
continue
if parent_node not in candidates:
candidates[parent_node] = self.score_node(parent_node)
ordered.append(parent_node)
if grand_parent_node is not None and grand_parent_node not in candidates:
candidates[grand_parent_node] = self.score_node(grand_parent_node)
ordered.append(grand_parent_node)
content_score = 1
content_score += len(inner_text.split(','))
content_score += min((inner_text_len / 100), 3)
#if elem not in candidates:
# candidates[elem] = self.score_node(elem)
#WTF? candidates[elem]['content_score'] += content_score
candidates[parent_node]['content_score'] += content_score
if grand_parent_node is not None:
candidates[grand_parent_node]['content_score'] += content_score / 2.0
# Scale the final candidates score based on link density. Good content should have a
# relatively small link density (5% or less) and be mostly unaffected by this operation.
for elem in ordered:
candidate = candidates[elem]
ld = self.get_link_density(elem)
score = candidate['content_score']
self.debug("Candid: %6.3f %s link density %.3f -> %6.3f" % (score, describe(elem), ld, score*(1-ld)))
candidate['content_score'] *= (1 - ld)
return candidates
def class_weight(self, e):
weight = 0
if e.get('class', None):
if REGEXES['negativeRe'].search(e.get('class')):
weight -= 25
if REGEXES['positiveRe'].search(e.get('class')):
weight += 25
if e.get('id', None):
if REGEXES['negativeRe'].search(e.get('id')):
weight -= 25
if REGEXES['positiveRe'].search(e.get('id')):
weight += 25
return weight
def score_node(self, elem):
content_score = self.class_weight(elem)
name = elem.tag.lower()
if name == "div":
content_score += 5
elif name in ["pre", "td", "blockquote"]:
content_score += 3
elif name in ["address", "ol", "ul", "dl", "dd", "dt", "li", "form"]:
content_score -= 3
elif name in ["h1", "h2", "h3", "h4", "h5", "h6", "th"]:
content_score -= 5
return {
'content_score': content_score,
'elem': elem
}
def debug(self, *a):
#if self.options['debug']:
logging.debug(*a)
def remove_unlikely_candidates(self):
for elem in self.html.iter():
s = "%s %s" % (elem.get('class', ''), elem.get('id', ''))
#self.debug(s)
if REGEXES['unlikelyCandidatesRe'].search(s) and (not REGEXES['okMaybeItsACandidateRe'].search(s)) and elem.tag != 'body':
self.debug("Removing unlikely candidate - %s" % describe(elem))
elem.drop_tree()
def transform_misused_divs_into_paragraphs(self):
for elem in self.tags(self.html, 'div'):
# transform
s that do not contain other block elements into
s
if not REGEXES['divToPElementsRe'].search(unicode(''.join(map(tostring, list(elem))))):
#self.debug("Altering %s to p" % (describe(elem)))
elem.tag = "p"
#print "Fixed element "+describe(elem)
for elem in self.tags(self.html, 'div'):
if elem.text and elem.text.strip():
p = fragment_fromstring('
')
p.text = elem.text
elem.text = None
elem.insert(0, p)
#print "Appended "+tounicode(p)+" to "+describe(elem)
for pos, child in reversed(list(enumerate(elem))):
if child.tail and child.tail.strip():
p = fragment_fromstring('')
p.text = child.tail
child.tail = None
elem.insert(pos + 1, p)
#print "Inserted "+tounicode(p)+" to "+describe(elem)
if child.tag == 'br':
#print 'Dropped at '+describe(elem)
child.drop_tree()
def tags(self, node, *tag_names):
for tag_name in tag_names:
for e in node.findall('.//%s' % tag_name):
yield e
def reverse_tags(self, node, *tag_names):
for tag_name in tag_names:
for e in reversed(node.findall('.//%s' % tag_name)):
yield e
def sanitize(self, node, candidates):
MIN_LEN = self.options.get('min_text_length', self.TEXT_LENGTH_THRESHOLD)
for header in self.tags(node, "h1", "h2", "h3", "h4", "h5", "h6"):
if self.class_weight(header) < 0 or self.get_link_density(header) > 0.33:
header.drop_tree()
for elem in self.tags(node, "form", "iframe", "textarea"):
elem.drop_tree()
allowed = {}
# Conditionally clean
s,
s, and
s
for el in self.reverse_tags(node, "table", "ul", "div"):
if el in allowed:
continue
weight = self.class_weight(el)
if el in candidates:
content_score = candidates[el]['content_score']
#print '!',el, '-> %6.3f' % content_score
else:
content_score = 0
tag = el.tag
if weight + content_score < 0:
self.debug("Cleaned %s with score %6.3f and weight %-3s" %
(describe(el), content_score, weight, ))
el.drop_tree()
elif el.text_content().count(",") < 10:
counts = {}
for kind in ['p', 'img', 'li', 'a', 'embed', 'input']:
counts[kind] = len(el.findall('.//%s' %kind))
counts["li"] -= 100
content_length = text_length(el) # Count the text length excluding any surrounding whitespace
link_density = self.get_link_density(el)
parent_node = el.getparent()
if parent_node is not None:
if parent_node in candidates:
content_score = candidates[parent_node]['content_score']
else:
content_score = 0
#if parent_node is not None:
#pweight = self.class_weight(parent_node) + content_score
#pname = describe(parent_node)
#else:
#pweight = 0
#pname = "no parent"
to_remove = False
reason = ""
#if el.tag == 'div' and counts["img"] >= 1:
# continue
if counts["p"] and counts["img"] > counts["p"]:
reason = "too many images (%s)" % counts["img"]
to_remove = True
elif counts["li"] > counts["p"] and tag != "ul" and tag != "ol":
reason = "more