Page MenuHomec4science

keyword_analyzer.py
No OneTemporary

File Metadata

Created
Wed, Dec 4, 17:06

keyword_analyzer.py

# -*- coding: utf-8 -*-
##
## This file is part of Invenio.
## Copyright (C) 2008, 2009, 2010, 2011, 2012, 2013, 2014 CERN.
##
## Invenio is free software; you can redistribute it and/or
## modify it under the terms of the GNU General Public License as
## published by the Free Software Foundation; either version 2 of the
## License, or (at your option) any later version.
##
## Invenio is distributed in the hope that it will be useful, but
## WITHOUT ANY WARRANTY; without even the implied warranty of
## MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
## General Public License for more details.
##
## You should have received a copy of the GNU General Public License
## along with Invenio; if not, write to the Free Software Foundation, Inc.,
## 59 Temple Place, Suite 330, Boston, MA 02111-1307, USA.
"""
BibClassify keyword analyser.
This module contains methods to extract keywords from texts. It provides 3
different methods for 3 different types of keywords: single keywords, composite
keywords and author keywords.
This module is STANDALONE safe
"""
from __future__ import print_function
import re
import time
import config as bconfig
log = bconfig.get_logger("bibclassify.keyword_analyzer")
_MAXIMUM_SEPARATOR_LENGTH = max([len(_separator)
for _separator in
bconfig.CFG_BIBCLASSIFY_VALID_SEPARATORS])
# XXX - rebuild this whole thing
def get_single_keywords(skw_db, fulltext):
"""Find single keywords in the fulltext
@var skw_db: list of KeywordToken objects
@var fulltext: string, which will be searched
@return : dictionary of matches in a format {
<keyword object>, [[position, position...], ],
..
}
"""
timer_start = time.clock()
# single keyword -> [spans]
records = []
for single_keyword in skw_db.values():
for regex in single_keyword.regex:
for match in regex.finditer(fulltext):
# Modify the right index to put it on the last letter
# of the word.
span = (match.span()[0], match.span()[1] - 1)
# FIXME: expensive!!!
# Remove the previous records contained by this span
records = [record for record in records
if not _contains_span(span, record[0])]
add = True
for previous_record in records:
if ((span, single_keyword) == previous_record or
_contains_span(previous_record[0], span)):
# Match is contained by a previous match.
add = False
break
if add:
records.append((span, single_keyword))
# TODO - change to the requested format (I will return to it later)
# List of single_keywords: {spans: single keyword}
single_keywords = {}
for span, single_keyword in records:
single_keywords.setdefault(single_keyword, [[]])
single_keywords[single_keyword][0].append(span)
log.info("Matching single keywords... %d keywords found "
"in %.1f sec." % (
len(single_keywords), time.clock() - timer_start),
)
return single_keywords
# XXX - rebuild this whole thing
def get_composite_keywords(ckw_db, fulltext, skw_spans):
"""Returns a list of composite keywords bound with the number of
occurrences found in the text string.
@var ckw_db: list of KewordToken objects (they are supposed to be composite ones)
@var fulltext: string to search in
@skw_spans: dictionary of already identified single keywords
@return : dictionary of matches in a format {
<keyword object>, [[position, position...], [info_about_matches] ],
..
}"""
timer_start = time.clock()
# Build the list of composite candidates
ckw_out = {}
skw_as_components = []
for composite_keyword in ckw_db.values():
# Counters for the composite keyword. First count is for the
# number of occurrences in the whole document and second count
# is for the human defined keywords.
ckw_count = 0
matched_spans = []
# First search in the fulltext using the regex pattern of the whole
# composite keyword (including the alternative labels)
for regex in composite_keyword.regex:
for match in regex.finditer(fulltext):
span = list(match.span())
span[1] -= 1
span = tuple(span)
if not span in matched_spans:
ckw_count += 1
matched_spans.append(span)
# Get the single keywords locations.
try:
components = composite_keyword.compositeof
except AttributeError:
print(log.error("Cached ontology is corrupted. Please "
"remove the cached ontology in your temporary file."))
raise Exception('Cached ontology is corrupted')
spans = []
try:
spans = [skw_spans[component][0] for component in components]
except KeyError:
# Some of the keyword components are not to be found in the text.
# Therefore we cannot continue because the match is incomplete.
continue
ckw_spans = []
for index in range(len(spans) - 1):
len_ckw = len(ckw_spans)
if ckw_spans: # cause ckw_spans include the previous
previous_spans = ckw_spans
else:
previous_spans = spans[index]
for new_span in [(span0, colmd1) for span0 in previous_spans
for colmd1 in spans[index + 1]]:
span = _get_ckw_span(fulltext, new_span)
if span is not None:
ckw_spans.append(span)
# the spans must be overlapping to be included
if index > 0 and ckw_spans:
_ckw_spans = []
for _span in ckw_spans[len_ckw:]: # new spans
for _colmd2 in ckw_spans[:len_ckw]:
s = _span_overlapping(_span, _colmd2)
if s:
_ckw_spans.append(s)
ckw_spans = _ckw_spans
for span in [span for span in ckw_spans
if not span in matched_spans]:
ckw_count += 1
matched_spans.append(span)
if ckw_count:
# Gather the component counts.
component_counts = []
for component in components:
skw_as_components.append(component)
# Get the single keyword count.
try:
component_counts.append(len(skw_spans[component][0]))
except KeyError:
component_counts.append(0)
# Store the composite keyword
ckw_out[composite_keyword] = [matched_spans, component_counts]
# Remove the single keywords that appear as components from the list
# of single keywords.
for skw in skw_as_components:
try:
del skw_spans[skw]
except KeyError:
pass
# Remove the composite keywords that are fully present in
# longer composite keywords
_ckw_base = filter(lambda x: len(x.compositeof) == 2, ckw_out.keys())
_ckw_extended = sorted(
filter(lambda x: len(x.compositeof) > 2, ckw_out.keys()),
key=lambda x: len(x.compositeof))
if _ckw_extended:
max_len = len(_ckw_extended[-1].compositeof)
candidates = []
for kw1 in _ckw_base:
s1 = set(kw1.compositeof)
for kw2 in _ckw_extended:
s2 = set(kw2.compositeof)
if s1.issubset(s2):
candidates.append((kw1, kw2))
#break # don't stop because this keyword may be
# partly contained by kw_x and kw_y
for i in range(len(_ckw_extended)):
kw1 = _ckw_extended[i]
s1 = set(kw1.compositeof)
for ii in range(i + 1, len(_ckw_extended)):
kw2 = _ckw_extended[ii]
s2 = set(kw2.compositeof)
if s1.issubset(s2):
candidates.append((kw1, kw2))
break
if candidates:
for kw1, kw2 in candidates:
match1 = ckw_out[kw1] # subset of the kw2
match2 = ckw_out[kw2]
positions1 = match1[0]
for pos1 in positions1:
for pos2 in match2[0]:
if _span_overlapping(pos1, pos2):
del positions1[positions1.index(pos1)]
if len(
positions1) == 0: # if we removed all the matches
del ckw_out[kw1] # also delete the keyword
break
log.info("Matching composite keywords... %d keywords found "
"in %.1f sec." % (len(ckw_out), time.clock() - timer_start),
)
return ckw_out
def get_author_keywords(skw_db, ckw_db, fulltext):
"""Finds out human defined keyowrds in a text string. Searches for
the string "Keywords:" and its declinations and matches the
following words."""
timer_start = time.clock()
out = {}
split_string = bconfig.CFG_BIBCLASSIFY_AUTHOR_KW_START.split(fulltext, 1)
if len(split_string) == 1:
log.info("Matching author keywords... no keywords marker found.")
return out
kw_string = split_string[1]
for regex in bconfig.CFG_BIBCLASSIFY_AUTHOR_KW_END:
parts = regex.split(kw_string, 1)
kw_string = parts[0]
# We separate the keywords.
author_keywords = bconfig.CFG_BIBCLASSIFY_AUTHOR_KW_SEPARATION.split(
kw_string)
log.info("Matching author keywords... %d keywords found in "
"%.1f sec." % (len(author_keywords), time.clock() - timer_start))
for kw in author_keywords:
# If the author keyword is an acronym with capital letters
# separated by points, remove the points.
if re.match('([A-Z].)+$', kw):
kw = kw.replace('.', '')
# First try with the keyword as such, then lower it.
kw_with_spaces = ' %s ' % kw
matching_skw = get_single_keywords(skw_db, kw_with_spaces)
matching_ckw = get_composite_keywords(ckw_db, kw_with_spaces,
matching_skw)
if matching_skw or matching_ckw:
out[kw] = (matching_skw, matching_ckw)
continue
lowkw = kw.lower()
matching_skw = get_single_keywords(skw_db, ' %s ' % lowkw)
matching_ckw = get_composite_keywords(ckw_db, ' %s ' % lowkw,
matching_skw)
out[kw] = (matching_skw, matching_ckw)
return out
def _get_ckw_span(fulltext, spans):
"""Returns the span of the composite keyword if it is valid. Returns
None otherwise."""
if spans[0] < spans[1]:
words = (spans[0], spans[1])
dist = spans[1][0] - spans[0][1]
else:
words = (spans[1], spans[0])
dist = spans[0][0] - spans[1][1]
if dist == 0:
# Two keywords are adjacent. We have a match.
return (min(words[0] + words[1]),
max(words[0] + words[1])) #FIXME: huh, this is a bug?! a sum???
elif dist <= _MAXIMUM_SEPARATOR_LENGTH:
separator = fulltext[words[0][1]:words[1][0] + 1]
# Check the separator.
if separator.strip() in bconfig.CFG_BIBCLASSIFY_VALID_SEPARATORS:
return (min(words[0] + words[1]), max(words[0] + words[1]))
# There is no inclusion.
return None
def _contains_span(span0, span1):
"""Return true if span0 contains span1, False otherwise."""
if (span0 == span1 or
span0[0] > span1[0] or
span0[1] < span1[1]):
return False
return True
def _span_overlapping(aspan, bspan):
# there are 6 posibilities, 2 are false
if bspan[0] >= aspan[0]:
if bspan[0] > aspan[1]:
return
else:
if aspan[0] > bspan[1]:
return
return (min(aspan[0], bspan[0]), max(aspan[1], bspan[1]))

Event Timeline