def _build_clusters(self, element):
"""
build candidate clusters according to element
:return:
"""
descendants_tree = defaultdict(list)
descendants = descendants_of_body(element)
for descendant in descendants:
# if one element does not have enough siblings, it can not become a child of candidate element
if descendant.number_of_siblings + 1 < self.min_number:
continue
# if min length is larger than specified max length, it can not become a child of candidate element
if descendant.a_descendants_group_text_min_length > self.max_length:
continue
# if max length is smaller than specified min length, it can not become a child of candidate element
if descendant.a_descendants_group_text_max_length < self.min_length:
continue
# descendant element must have same siblings which their similarity should not below similarity_threshold
if descendant.similarity_with_siblings < self.similarity_threshold:
continue
descendants_tree[descendant.parent_selector].append(descendant)
descendants_tree = dict(descendants_tree)
# cut tree, remove parent block
selectors = sorted(list(descendants_tree.keys()))
last_selector = None
for selector in selectors[::-1]:
# if later selector
if last_selector and selector and last_selector.startswith(
selector):
del descendants_tree[selector]
last_selector = selector
clusters = cluster_dict(descendants_tree)
return clusters
def process(self, element: Element):
"""
extract content from html
:param element:
:return:
"""
# preprocess
preprocess4content(element)
# start to evaluate every child element
element_infos = []
descendants = descendants_of_body(element)
# get std of density_of_text among all elements
density_of_text = [
descendant.density_of_text for descendant in descendants
]
density_of_text_std = np.std(density_of_text, ddof=1)
# get density_score of every element
for descendant in descendants:
score = np.log(density_of_text_std) * \
descendant.density_of_text * \
np.log10(descendant.number_of_p_descendants + 2) * \
np.log(descendant.density_of_punctuation)
descendant.density_score = score
# sort element info by density_score
descendants = sorted(descendants,
key=lambda x: x.density_score,
reverse=True)
descendant_first = descendants[0] if descendants else None
if descendant_first is None:
return None
paragraphs = descendant_first.xpath('.//p//text()')
paragraphs = [
paragraph.strip() if paragraph else '' for paragraph in paragraphs
]
paragraphs = list(filter(lambda x: x, paragraphs))
text = '\n'.join(paragraphs)
text = text.strip()
return text
