def getInterestingLabel(self, treeNode):
"""
Finds 'interesting things' such as inversions, shifters,
etc.
"""
if len(treeNode) < 2:
return None
if self.granularity == ma_util.GRANULARITY_COARSE:
raise NotImplementedError(
'getInterestingLabels does not handle coarse labels!')
catP = ma_util.sen(treeNode.node, self.granularity)
catL = ma_util.sen(treeNode[0].node, self.granularity)
catR = ma_util.sen(treeNode[1].node, self.granularity)
if catP == catL and catP == catR:
return "ID"
cat_max = max(catL, catR)
cat_min = min(catL, catR)
cat_avg = (catL + catR) / 2.0
# cat_miv is low if max sentiment is very POS
cat_miv = ma_util.VERY_POS - cat_max
# cat_mav is low if min sentiment is very POS
cat_mav = ma_util.VERY_POS - cat_min
if catP <= cat_max and catP >= cat_min:
r_label = 'AVG'
# is parent less positive than most positive child
# and more positive than least positive?
elif catP <= cat_mav and catP >= cat_miv:
r_label = 'INV'
elif catL == ma_util.FINE_NEU and ((catP - ma_util.FINE_NEU)
* (catR - ma_util.FINE_NEU) < 0):
#r_label = 'INV/L'
r_label = 'INV'
elif catR == ma_util.FINE_NEU and ((catP - ma_util.FINE_NEU)
* (catL - ma_util.FINE_NEU) < 0):
#r_label = 'INV/R'
r_label = 'INV'
elif (catP - ma_util.FINE_NEU) * (cat_avg - ma_util.FINE_NEU) > 0:
# (3 (4 wundervollen) (2 Szenen)) would match
# here, but other rules fire first
# Also not sure if there is intensification happening there?
r_label = 'INT'
else:
#r_label = '???'
r_label = "MWE"
return r_label
def getProduction(self, treeNode):
"""
>>> pe = ProdEvaluator()
>>> from nltk.tree import Tree
>>> t = Tree(3, [Tree(3, "good"), Tree(2, "movie")])
>>> pe.getProduction(t)
(3,3,2)
>>> t = Tree(3, "good")
>>> pe.getProduction(t)
None
"""
# If the tree is binary, then its children are other tree instances
# unary nodes only have for terminals and associated sentiment label
if len(treeNode) < 2:
return None
top = ma_util.sen(treeNode.node, self.granularity)
c0 = ma_util.sen(treeNode[0].node, self.granularity)
c1 = ma_util.sen(treeNode[1].node, self.granularity)
return (top, c0, c1)
def extract_projection_sentiment(self, tigerObj, tigerSentence):
"""Extracts projection feature.
@param tigerObj {TigerXML} Full parse file
@param tigerSentence {some XML Element} XML node of current sentence
@returns {list} Sentiment Values for each Span NEG, NEU, POS
"""
data = []
for node in tigerObj.preOrder(tigerSentence, forceSentiment=True):
data.append(ma_util.sen(node[1], self.granularity))
return data
def extract_gold_sentiment(self, goldSentence, extractLength=False):
"""Extracts gold label.
@param goldSentence {PTB tree} Parse tree with sentiment annotation
@returns {list} List of labels
"""
data = []
for goldNode in ma_util.walkTree(goldSentence):
label = ma_util.sen(goldNode.node, self.granularity)
if extractLength:
data.append((label, len(goldNode.leaves())))
else:
data.append(label)
return data