def evaluate_sentence_similarity(self, dataset_name="MSRvid", metric = "wpath_graph", relatedness=True, save_results = False, database="wikidata"):
concepts, cc, texts = get_ideas_in_format(dataset_name, database=database)
KG = DAC(concepts=concepts, dataset=dataset_name, relatedness=relatedness, database=database)
if(KG.graph.__len__()==0):
print("start building knowledge graph")
KG.build_nx_graph()
ConSim = ConceptSimilarity(KG)
sim_M = ConSim.similarityMatrix(lcs_pref_value="freq1", metric=metric)
WMD = WordMoversSimilarity(sim_M, KG._concepts)
sen_pairs, human_sim = self._dataset.load_sentence_pairs_and_similarities(dataset_name)
sim_values = []
map_sen2bow = dict(zip(texts, [[c["id"] for c in bow] for bow in cc]))
pg, total_len = 0 , len(sen_pairs)
remove_index = []
for sen1, sen2 in sen_pairs:
show_progression(pg, total_len)
bow1, bow2 = list(set(map_sen2bow[sen1]) & set(KG._concepts)), list(set(map_sen2bow[sen2]) & set(KG._concepts))
sim_value = WMD.word_mover_distance(bow1, bow2)
if sim_value is None:
print(sen1, sen2)
remove_index.append(pg)
else:
sim_values.append(sim_value)
pg = pg+1
human_sim = np.delete(human_sim, remove_index)
cor = pearsonr(sim_values, human_sim)[0]
if save_results:
results = list(zip([round(x, 3) for x in sim_values], sen_pairs))
self._dataset.save_dataset(dict(zip(("correlation", "similarities"),(cor, results))), dataset_name+"_"+metric)
return cor
def global_secondorder_freq(self, write_value="freq1"):
i, num_nodes = 0, self.graph.__len__()
for n in self.graph.nodes:
i = i + 1
show_progression(i, num_nodes)
try:
num1 = sparql_request(
sql.query_babelnet_number_of(n))[0]["count"]["value"]
self.graph.nodes[n][write_value] = int(num1)
except:
print("timeout babelnet", n)
self.graph.nodes[n][write_value] = 6000000
self.write_to_file()
def build_nx_graph(self, query=sql.query_ancestors):
start_t = time.time()
concepts = self._init_concepts
nxKG = nx.DiGraph()
i, concepts_len = 0, len(concepts)
for con in concepts:
keys = [con]
i = i + 1
show_progression(i, concepts_len)
if self._relatedness:
keys += self.add_edges_for_concept(
query(con, sql.relation_prop), nxKG)
while len(keys) > 0:
key = keys.pop()
keys += self.add_edges_for_concept(query(key), nxKG)
self.graph = nxKG
self.init_key2pos()
self.write_to_file()
return print("building the Graph from %s took: %s seconds." %
(self._database, (time.time() - start_t)))
def build_nx_graph(self):
if self._database != "babelnet":
return print("Function only for babelnet")
start_t = time.time()
concepts = self._init_concepts
nxKG = nx.DiGraph()
i, concepts_len = 0, len(concepts)
for con in concepts:
keys = [con]
i = i + 1
show_progression(i, concepts_len)
while len(keys) > 0:
key = keys.pop()
keys += self.add_edges_for_concept(sql.babelnet_paths2top(key),
nxKG,
split_delimiter="/s",
prefix="bn:")
self.graph = nxKG
self.init_key2pos()
self.write_to_file()
return print("building the Graph from %s took: %s seconds." %
(self._database, (time.time() - start_t)))
def information_content(concepts):
IC = []
for c, i in concepts:
show_progression(i, len(concepts))
IC.append(sparql_request(sql.query_freq_wikidata(c, 2), "wikidata"))
return [c[0]["count"]["value"] if c != None else None for c in IC]