def add(self_node: Synset,
parent_list: List[Synset],
g: nx.DiGraph,
parent: bool = True):
for n in parent_list:
g.add_node(n[0].name())
if parent:
g.add_edge(self_node.name(), n[0].name())
else:
g.add_edge(n[0].name(), self_node.name())
return g
def simplified_lesk(word: str, sentence: str) -> Synset:
"""
Computes the max_overlap to understand what is the best sense
:param word: word to dissmbiguate
:param sentence: sentence in which word appears
:return: Synset that maximizes the overlap
"""
synsets = wordnet.synsets(word)
try:
lemmatizer = WordNetLemmatizer()
best_sense = wordnet.synsets(word)[0]
max_overlap = 0
context = set(lemmatizer.lemmatize(word)for word in sentence.split(" "))
for sense in synsets:
signature = set(lemmatizer.lemmatize(word)for word in sense.definition().split(" "))
for example in sense.examples():
signature.union(set(lemmatizer.lemmatize(word)for word in example.split(" ")))
overlap = len(signature.intersection(context))
if overlap > max_overlap:
max_overlap = overlap
best_sense = sense
return best_sense
except:
return Synset(None)
def get_parent(synset: Synset):
"""
Returns one of the parents of the synset.
:param synset: The synset to obtain the parent from
:return: One of the parents of the synset
"""
return random.choice(synset.hypernyms())
def get_parents(synset: Synset):
"""
Returns all parents of the synset (hypernyms).
:param synset: The synset to obtain the parent from
:return: List of the parents of the synset
"""
return synset.hypernyms()
def get_synset_id(synset: Synset):
"""
Get the corresponding synset id of the synset.
:param synset: The synset to extract the id from
:return: The corresponding synset id
"""
sid = "n{}".format(str(synset.offset()).zfill(8))
return sid
def get_grandparents(synset: Synset):
"""
Returns all grandparents of the synset.
:param synset: The synset to obtain the grandparents from
:return: The grandparents of the synset
"""
grandparents = []
for parent in synset.hypernyms():
grandparents.extend(parent.hypernyms())
return grandparents
def extended_context_lesk(word: str, sentence: str) -> Synset:
"""
Computes the max_overlap to understand what is the best sense, using hypernyms and hyponyms
:param word: the word to be disambiguated
:param sentence: input sentence which contains param 'word'
:return: best_sense, which is a Wordnet Synset, for param 'word'
"""
stopwords_set = set(stopwords.words('english'))
synsets = wordnet.synsets(word)
try:
lemmatizer = WordNetLemmatizer()
best_sense = wordnet.synsets(word)[0]
max_overlap = 0
context = set(lemmatizer.lemmatize(word)for word in sentence.split(" "))
for sense in synsets:
signature = set(lemmatizer.lemmatize(word)for word in sense.definition().split(" "))
for example in sense.examples():
signature.union(set(lemmatizer.lemmatize(word)for word in example.split(" ")))
for hypernym in sense.hypernyms():
signature = signature.union(set(lemmatizer.lemmatize(word)for word in hypernym.definition().split(" ")))
for hyponym in sense.hyponyms():
signature = signature.union(set(lemmatizer.lemmatize(word)for word in hyponym.definition().split(" ")))
signature.difference(stopwords_set)
overlap = len(signature.intersection(context))
if overlap > max_overlap:
max_overlap = overlap
best_sense = sense
return best_sense
except:
return Synset(None)
def lin(self, s1: Synset, s2: Synset) -> float:
return s1.lin_similarity(s2, self.brown_ic)
def res(self, s1: Synset, s2: Synset) -> float:
return s1.res_similarity(s2, self.brown_ic)
def wup(self, s1: Synset, s2: Synset) -> float:
return s1.wup_similarity(s2)
def lch(self, s1: Synset, s2: Synset) -> float:
return s1.lch_similarity(s2)
def path(self, s1: Synset, s2: Synset) -> float:
return s1.path_similarity(s2)
def get_synset_frequency(synset: Synset):
"""
Get synset frequency in Brown
"""
return sum([lemma.count() for lemma in synset.lemmas()])
def meronym_ancestors(self, i):
return get_ancestors(self.synset(i), Synset.member_meronyms())
def lin(self, s1: Synset, s2: Synset) -> float:
try:
return s1.lin_similarity(s2, self.brown_ic)
except WordNetError:
return NAN
def lch(self, s1: Synset, s2: Synset) -> float:
try:
return s1.lch_similarity(s2)
except WordNetError:
return NAN
def get_graph(start: Synset, relation: Callable[[Synset], List[Synset]]):
""" Get the graph in the form of recurssive lists formed by doing
depth-first walk starting from start and following edges relation"""
return start.tree(relation)
def lowest_common_hypernyms(self,
synset,
other,
simulate_root=False,
use_min_depth=False):
'''
-- NOTE: THIS CODE IS COPIED FROM NLTK3 --
Get a list of lowest synset(s) that both synsets have as a hypernym.
When `use_min_depth == False` this means that the synset which
appears as a hypernym of both `self` and `other` with the lowest
maximum depth is returned or if there are multiple such synsets at
the same depth they are all returned
However, if `use_min_depth == True` then the synset(s) which has/have
the lowest minimum depth and appear(s) in both paths is/are returned.
By setting the use_min_depth flag to True, the behavior of NLTK2 can
be preserved. This was changed in NLTK3 to give more accurate results
in a small set of cases, generally with synsets concerning people.
(eg: 'chef.n.01', 'fireman.n.01', etc.)
This method is an implementation of Ted Pedersen's "Lowest Common
Subsumer" method from the Perl Wordnet module. It can return either
"self" or "other" if they are a hypernym of the other.
:type other: Synset
:param other: other input synset
:type simulate_root: bool
:param simulate_root: The various verb taxonomies do not
share a single root which disallows this metric from working for
synsets that are not connected. This flag (False by default)
creates a fake root that connects all the taxonomies. Set it
to True to enable this behavior. For the noun taxonomy,
there is usually a default root except for WordNet version 1.6.
If you are using wordnet 1.6, a fake root will need to be added
for nouns as well.
:type use_min_depth: bool
:param use_min_depth: This setting mimics older (v2) behavior of NLTK
wordnet. If True, will use the min_depth function to calculate
the lowest common hypernyms. This is known to give strange
results for some synset pairs (eg: 'chef.n.01', 'fireman.n.01')
but is retained for backwards compatibility
:return: The synsets that are the lowest common hypernyms of both
synsets
'''
fake_synset = Synset(None)
fake_synset._name = '*ROOT*'
fake_synset.hypernyms = lambda: []
fake_synset.instance_hypernyms = lambda: []
if simulate_root:
self_hypernyms = chain(synset._iter_hypernym_lists(),
[[fake_synset]])
other_hypernyms = chain(other._iter_hypernym_lists(),
[[fake_synset]])
else:
self_hypernyms = synset._iter_hypernym_lists()
other_hypernyms = other._iter_hypernym_lists()
synsets = set(s for synsets in self_hypernyms for s in synsets)
others = set(s for synsets in other_hypernyms for s in synsets)
if self.core_taxonomy is not None:
synsets.intersection_update(
map(lambda syn: wordnet.synset(syn), self.known_concepts))
others.intersection_update(
map(lambda syn: wordnet.synset(syn), self.known_concepts))
synsets.intersection_update(others)
try:
if use_min_depth:
max_depth = max(s.min_depth() for s in synsets)
unsorted_lch = [
s for s in synsets if s.min_depth() == max_depth
]
else:
max_depth = max(s.max_depth() for s in synsets)
unsorted_lch = [
s for s in synsets if s.max_depth() == max_depth
]
return sorted(unsorted_lch)
except ValueError:
return []