def build_content_sim_relation_text_lsa(network, signatures):
def get_nid_gen(signatures):
for nid, sig in signatures:
yield nid
docs = []
for nid, e in signatures:
docs.append(' '.join(e))
# this may become redundant if we exploit the store characteristics
tfidf = da.get_tfidf_docs(docs)
print("TF-IDF shape before LSA: " + str(tfidf.shape))
st = time.time()
tfidf = lsa_dimensionality_reduction(tfidf)
et = time.time()
print("TF-IDF shape after LSA: " + str(tfidf.shape))
print("Time to compute LSA: {0}".format(str(et - st)))
lsh_projections = RandomBinaryProjections('default', 10000)
#lsh_projections = RandomDiscretizedProjections('rnddiscretized', 1000, 2)
nid_gen = get_nid_gen(signatures) # to preserve the order nid -> signature
text_engine = index_in_text_engine(nid_gen,
tfidf,
lsh_projections,
tfidf_is_dense=True)
nid_gen = get_nid_gen(signatures) # to preserve the order nid -> signature
create_sim_graph_text(nid_gen,
network,
text_engine,
tfidf,
Relation.CONTENT_SIM,
tfidf_is_dense=True)
def build_entity_sim_relation(network, fields, entities):
docs = []
for e in entities:
if e != "": # Append only non-empty documents
docs.append(e)
print(str(docs))
if len(docs) > 0: # If documents are empty, then skip this step; not entity similarity will be found
tfidf = da.get_tfidf_docs(docs)
text_engine = index_in_text_engine(
fields, tfidf, rbp) # rbp the global variable
create_sim_graph_text(network, text_engine, fields,
tfidf, Relation.ENTITY_SIM)
def build_schema_sim_relation(network):
def connect(nid1, nid2, score):
network.add_relation(nid1, nid2, Relation.SCHEMA_SIM, score)
st = time.time()
docs = []
for (_, _, field_name, _) in network.iterate_values():
docs.append(field_name)
tfidf = da.get_tfidf_docs(docs)
et = time.time()
print("Time to create docs and TF-IDF: ")
print("Create docs and TF-IDF: {0}".format(str(et - st)))
nid_gen = network.iterate_ids()
num_features = tfidf.shape[1]
new_index_engine = LSHRandomProjectionsIndex(num_features)
# Index vectors in engine
st = time.time()
row_idx = 0
for key in nid_gen:
sparse_row = tfidf.getrow(row_idx)
dense_row = sparse_row.todense()
array = dense_row.A[0]
row_idx += 1
new_index_engine.index(array, key)
et = time.time()
print("Total index text: " + str((et - st)))
# Create schema_sim links
nid_gen = network.iterate_ids()
st = time.time()
row_idx = 0
for nid in nid_gen:
sparse_row = tfidf.getrow(row_idx)
dense_row = sparse_row.todense()
array = dense_row.A[0]
row_idx += 1
N = new_index_engine.query(array)
if len(N) > 1:
for n in N:
(data, key, value) = n
if nid != key:
connect(nid, key, value)
et = time.time()
print("Create graph schema: {0}".format(str(et - st)))
return new_index_engine
def build_schema_sim_relation_lsa(network, fields):
docs = []
for (nid, sn, fn, _, _) in fields:
docs.append(fn)
tfidf = da.get_tfidf_docs(docs)
print("tfidf shape before LSA: " + str(tfidf.shape))
tfidf = lsa_dimensionality_reduction(tfidf)
print("tfidf shape after LSA: " + str(tfidf.shape))
text_engine = index_in_text_engine(
fields, tfidf, rbp, tfidf_is_dense=True) # rbp the global variable
create_sim_graph_text(network, text_engine, fields,
tfidf, Relation.SCHEMA_SIM, tfidf_is_dense=True)
def build_content_sim_relation_text(network, signatures):
def get_nid_gen(signatures):
for nid, sig in signatures:
yield nid
docs = []
for nid, e in signatures:
docs.append(' '.join(e))
# this may become redundant if we exploit the store characteristics
tfidf = da.get_tfidf_docs(docs)
# rbp = RandomBinaryProjections('default', 1000)
lsh_projections = RandomDiscretizedProjections('rnddiscretized', 1000, 2)
nid_gen = get_nid_gen(signatures)
text_engine = index_in_text_engine(nid_gen, tfidf, lsh_projections)
nid_gen = get_nid_gen(signatures)
create_sim_graph_text(nid_gen, network, text_engine, tfidf,
Relation.CONTENT_SIM)
def __find_relation_class_matchings(self):
# Retrieve relation names
st = time.time()
docs = []
names = []
seen_sources = []
for (_, source_name, _, _) in self.network.iterate_values():
if source_name not in seen_sources:
seen_sources.append(source_name) # seen already
source_name = source_name.replace('-', ' ')
source_name = source_name.replace('_', ' ')
source_name = source_name.lower()
docs.append(source_name)
names.append(('relation', source_name))
# Retrieve class names
for kr_item, kr_handler in self.kr_handlers.items():
all_classes = kr_handler.classes()
for cl in all_classes:
cl = cl.replace('-', ' ')
cl = cl.replace('_', ' ')
cl = cl.lower()
docs.append(cl)
names.append(('class', cl))
tfidf = da.get_tfidf_docs(docs)
et = time.time()
print("Time to create docs and TF-IDF: ")
print("Create docs and TF-IDF: {0}".format(str(et - st)))
num_features = tfidf.shape[1]
new_index_engine = LSHRandomProjectionsIndex(num_features,
projection_count=7)
# N2 method
"""
clean_matchings = []
for i in range(len(docs)):
for j in range(len(docs)):
sparse_row = tfidf.getrow(i)
dense_row = sparse_row.todense()
array_i = dense_row.A[0]
sparse_row = tfidf.getrow(j)
dense_row = sparse_row.todense()
array_j = dense_row.A[0]
sim = np.dot(array_i, array_j.T)
if sim > 0.5:
if names[i][0] != names[j][0]:
match = names[i][1], names[j][1]
clean_matchings.append(match)
return clean_matchings
"""
# Index vectors in engine
st = time.time()
for idx in range(len(docs)):
sparse_row = tfidf.getrow(idx)
dense_row = sparse_row.todense()
array = dense_row.A[0]
new_index_engine.index(array, idx)
et = time.time()
print("Total index text: " + str((et - st)))
# Now query for similar ones:
raw_matchings = defaultdict(list)
for idx in range(len(docs)):
sparse_row = tfidf.getrow(idx)
dense_row = sparse_row.todense()
array = dense_row.A[0]
N = new_index_engine.query(array)
if len(N) > 1:
for n in N:
(data, key, value) = n
raw_matchings[idx].append(key)
et = time.time()
print("Find raw matches: {0}".format(str(et - st)))
# Filter matches so that only relation-class ones appear
clean_matchings = []
for key, values in raw_matchings.items():
key_kind = names[key][0]
for v in values:
v_kind = names[v][0]
if v_kind != key_kind:
match = (names[key][1], names[v][1])
clean_matchings.append(match)
return clean_matchings
