def _negation_calc(self, tags, window):
'''
for a list of tokens, calculate array of negated words based on a negation detection
algorithm (NegEx in our case).
returns arran vNEG containing [0,1] for each index of token on original tags list, indicating negation.
'''
vNEG = negdetect.getNegationArray(tags, window)
return vNEG
def _negation_calc(self, tags, window):
'''
for a list of tokens, calculate array of negated words based on a negation detection
algorithm (NegEx in our case).
returns arran vNEG containing [0,1] for each index of token on original tags list, indicating negation.
'''
vNEG = negdetect.getNegationArray(tags, window)
return vNEG
def docSentiScore(L, Doc, aflag, vflag, rflag, nflag, negflag, negwindow=5,
w = lambda x,y,z:x,
scoringmethod = SCOREALL):
'''
docSentiScore - Computes document sentiment score for POS-tagged Doc using sentiment lexicon L.
The following mandatory flags specify which POS tags to take into account:
- aflag
- vflag
- rflag (adverbs)
- nflag
In addition, negflag is a boolean indicating whether to use negation detection, with window scope negwindow.
w(x,i,N) is a weight adjustment function based on word position within the document.
scoringmethod is a list of non-exlcusive parameters used to switch on/off scoring features.
Returns tuple (posscore, negscore, doc) containing final document scores and annotated document.
'''
# 0. Initialize
tags = Doc.split()
annotatedTags = []
doclen = len(tags)
i=0
postotal=0
negtotal=0
notfoundcounter=0
foundcounter=0
negcount=0
# Setup stem preprocessing for verbs
wnl = nltk.stem.WordNetLemmatizer()
# 1. Negation detection pre-processing - return an array w/ position of negated terms
vNEG = negdetect.getNegationArray(tags, negwindow)
#
# 2. Scan for scores for each POS
# After POS-tagging a term will appear as either term/POS or term_POS
# We assume such weirdnesses will not naturally occur on plain text.
#
posindex=0
negindex=1
tagList = []
for tagword in tags:
i+=1
scoretuple = (0,0)
tagfound = False
tagseparator = ''
# Adjectives
if aflag==True and re.search('[_/](JJ|JJ.)$',tagword):
tagfound=True
tagseparator = tagword[ re.search('[_/](JJ|JJ.)$',tagword).start() ]
thisterm = tagword.split( tagseparator )[0]
scoretuple = L.getadjective( thisterm )
# Verbs (VBP / VBD/ etc...)
if vflag==True and re.search('[_/](VB|VB.)$',tagword) :
tagfound=True
tagseparator = tagword[ re.search('[_/](VB|VB.)$',tagword).start() ]
thisterm = tagword.split( tagseparator )[0]
thisterm = wnl.lemmatize(thisterm, pos='v')
scoretuple = L.getverb( thisterm )
# Adverbs
if rflag==True and re.search('[_/]RB$',tagword) :
tagfound=True
tagseparator = tagword[ re.search('[_/]RB$',tagword).start() ]
thisterm = tagword.split( tagseparator )[0]
scoretuple = L.getadverb(thisterm)
# Nouns
if nflag==True and re.search('[_/]NN$',tagword) :
tagfound=True
tagseparator = tagword[ re.search('[_/]NN$',tagword).start() ]
thisterm = tagword.split( tagseparator )[0]
scoretuple = L.getnoun(thisterm)
# Process negation detection
if negflag==True:
posindex=vNEG[i-1]
negindex=(1+vNEG[i-1])%2
#
# Add to total with weight score
#
posval = 0.0
negval = 0.0
if tagfound and \
(
(scoringmethod == SCOREALL) or
(scoringmethod == SCOREWITHFREQ) or
(scoringmethod == SCOREONCE and (not tagword in tagList)) or
(scoringmethod == SCOREWITHSTOP and (not objectiveWords.is_stop(thisterm)) )
):
if (scoringmethod in [ SCOREWITHFREQ, SCOREWITHSTOP ] ):
# Scoring with frequency information
# Frequency is a real valued at 0.0-1.0. We calculate sqrt function so that the value grows faster even for numbers close to 0
posval += w(scoretuple[posindex],i,doclen) * (1.0 - math.sqrt(L.get_freq( thisterm )) )
negval += w(scoretuple[negindex],i,doclen) * (1.0 - math.sqrt(L.get_freq( thisterm )) )
else:
# Just plain scoring from lexicon - add
posval += w(scoretuple[posindex],i,doclen)
negval += w(scoretuple[negindex],i,doclen)
postotal += posval
negtotal += negval
# Found a tag - increase counters and add tag to list
if tagfound:
tagList.append( tagword )
foundcounter += 1.0
if negflag==True and vNEG[i-1]==1:
negcount += 1
if scoretuple == (0,0):
notfoundcounter += 1.0
# add this tag back to annotated version
if tagfound:
if negflag:
negtag = str(vNEG[i-1])
else:
negtag = 'NONEG'
annotatedTags.append( tagword + '##NEGAT:' + negtag + '##POS:' + str(posval) + '##NEG:' + str(negval) )
else:
annotatedTags.append( tagword )
# Completed scan
# Add negated words to negative score (Potts 2011)
# alpha is a scaling factor based on how much of the document has been negated
if foundcounter > 0.0:
alpha = (negcount)/(foundcounter)
else:
alpha = 0.0
##negtotal += negcount * ( 0.5 * (1 - alpha) )
#negtotal += negcount * 0.3
if (foundcounter - notfoundcounter) > 0.0:
# resulting scores are the normalized proportions of all *scored* terms (ignoring neutrals/unknowns)
resultpos = float(postotal)/(foundcounter - notfoundcounter )
resultneg = float(negtotal)/(foundcounter - notfoundcounter )
else:
resultpos = 0.0
resultneg = 0.0
return (resultpos, resultneg, ' '.join(annotatedTags))
def docSentiScore(L,
Doc,
aflag,
vflag,
rflag,
nflag,
negflag,
negwindow=5,
w=lambda x, y, z: x,
scoringmethod=SCOREALL):
'''
docSentiScore - Computes document sentiment score for POS-tagged Doc using sentiment lexicon L.
The following mandatory flags specify which POS tags to take into account:
- aflag
- vflag
- rflag (adverbs)
- nflag
In addition, negflag is a boolean indicating whether to use negation detection, with window scope negwindow.
w(x,i,N) is a weight adjustment function based on word position within the document.
scoringmethod is a list of non-exlcusive parameters used to switch on/off scoring features.
Returns tuple (posscore, negscore, doc) containing final document scores and annotated document.
'''
# 0. Initialize
tags = Doc.split()
annotatedTags = []
doclen = len(tags)
i = 0
postotal = 0
negtotal = 0
notfoundcounter = 0
foundcounter = 0
negcount = 0
# Setup stem preprocessing for verbs
wnl = nltk.stem.WordNetLemmatizer()
# 1. Negation detection pre-processing - return an array w/ position of negated terms
vNEG = negdetect.getNegationArray(tags, negwindow)
#
# 2. Scan for scores for each POS
# After POS-tagging a term will appear as either term/POS or term_POS
# We assume such weirdnesses will not naturally occur on plain text.
#
posindex = 0
negindex = 1
tagList = []
for tagword in tags:
i += 1
scoretuple = (0, 0)
tagfound = False
tagseparator = ''
# Adjectives
if aflag == True and re.search('[_/](JJ|JJ.)$', tagword):
tagfound = True
tagseparator = tagword[re.search('[_/](JJ|JJ.)$', tagword).start()]
thisterm = tagword.split(tagseparator)[0]
scoretuple = L.getadjective(thisterm)
# Verbs (VBP / VBD/ etc...)
if vflag == True and re.search('[_/](VB|VB.)$', tagword):
tagfound = True
tagseparator = tagword[re.search('[_/](VB|VB.)$', tagword).start()]
thisterm = tagword.split(tagseparator)[0]
thisterm = wnl.lemmatize(thisterm, pos='v')
scoretuple = L.getverb(thisterm)
# Adverbs
if rflag == True and re.search('[_/]RB$', tagword):
tagfound = True
tagseparator = tagword[re.search('[_/]RB$', tagword).start()]
thisterm = tagword.split(tagseparator)[0]
scoretuple = L.getadverb(thisterm)
# Nouns
if nflag == True and re.search('[_/]NN$', tagword):
tagfound = True
tagseparator = tagword[re.search('[_/]NN$', tagword).start()]
thisterm = tagword.split(tagseparator)[0]
scoretuple = L.getnoun(thisterm)
# Process negation detection
if negflag == True:
posindex = vNEG[i - 1]
negindex = (1 + vNEG[i - 1]) % 2
#
# Add to total with weight score
#
posval = 0.0
negval = 0.0
if tagfound and \
(
(scoringmethod == SCOREALL) or
(scoringmethod == SCOREWITHFREQ) or
(scoringmethod == SCOREONCE and (not tagword in tagList)) or
(scoringmethod == SCOREWITHSTOP and (not objectiveWords.is_stop(thisterm)) )
):
if (scoringmethod in [SCOREWITHFREQ, SCOREWITHSTOP]):
# Scoring with frequency information
# Frequency is a real valued at 0.0-1.0. We calculate sqrt function so that the value grows faster even for numbers close to 0
posval += w(scoretuple[posindex], i,
doclen) * (1.0 - math.sqrt(L.get_freq(thisterm)))
negval += w(scoretuple[negindex], i,
doclen) * (1.0 - math.sqrt(L.get_freq(thisterm)))
else:
# Just plain scoring from lexicon - add
posval += w(scoretuple[posindex], i, doclen)
negval += w(scoretuple[negindex], i, doclen)
postotal += posval
negtotal += negval
# Found a tag - increase counters and add tag to list
if tagfound:
tagList.append(tagword)
foundcounter += 1.0
if negflag == True and vNEG[i - 1] == 1:
negcount += 1
if scoretuple == (0, 0):
notfoundcounter += 1.0
# add this tag back to annotated version
if tagfound:
if negflag:
negtag = str(vNEG[i - 1])
else:
negtag = 'NONEG'
annotatedTags.append(tagword + '##NEGAT:' + negtag + '##POS:' +
str(posval) + '##NEG:' + str(negval))
else:
annotatedTags.append(tagword)
# Completed scan
# Add negated words to negative score (Potts 2011)
# alpha is a scaling factor based on how much of the document has been negated
if foundcounter > 0.0:
alpha = (negcount) / (foundcounter)
else:
alpha = 0.0
##negtotal += negcount * ( 0.5 * (1 - alpha) )
#negtotal += negcount * 0.3
if (foundcounter - notfoundcounter) > 0.0:
# resulting scores are the normalized proportions of all *scored* terms (ignoring neutrals/unknowns)
resultpos = float(postotal) / (foundcounter - notfoundcounter)
resultneg = float(negtotal) / (foundcounter - notfoundcounter)
else:
resultpos = 0.0
resultneg = 0.0
return (resultpos, resultneg, ' '.join(annotatedTags))
