def similarity(s1, s2):
s1 = process_line_msrp(s1)
s2 = process_line_msrp(s2)
# print "s1:", s1
# print "s2:", s2
s1_word = s1.split(" ")
s2_word = s2.split(" ")
union_s = list(set(s2_word).union(set(s1_word)))
# print "union_s:", union_s
n = len(union_s)
s1_vector = np.zeros(n)
s2_vector = np.zeros(n)
i = 0
for word in union_s:
if word in s1_word:
s1_vector[i] = 1
if word in s2_word:
s2_vector[i] = 1
i += 1
# print "s1_vector:", s1_vector
# print "s2_vector:", s2_vector
# print np.dot(s1_vector, s2_vector) / (np.linalg.norm(s1_vector) * np.linalg.norm(s2_vector))
matrix_wup = np.zeros((n, n))
i = 0
j = 0
for w1 in union_s:
for w2 in union_s:
if matrix_wup[i][j] > 0:
j += 1
continue
if i == j:
matrix_wup[i][j] = 1
j += 1
continue
w1 = w1.lower()
w2 = w2.lower()
if w1 == w2:
matrix_wup[i][j] = 1
j += 1
continue
if not w1.isalpha():
matrix_wup[i][j] = 0
j += 1
continue
if not w2.isalpha():
matrix_wup[i][j] = 0
j += 1
continue
w1_synsets = wn.synsets(w1)
if len(w1_synsets) == 0:
matrix_wup[i][j] = 0
j += 1
continue
w1_synset = w1_synsets[0]
w2_synsets = wn.synsets(w2)
if len(w2_synsets) == 0:
matrix_wup[i][j] = 0
j += 1
continue
w2_synset = w2_synsets[0]
if w1_synset.pos() != w2_synset.pos():
matrix_wup[i][j] = 0
j += 1
continue
wup_sim = w1_synset.wup_similarity(w2_synset)
if wup_sim < 0.8:
wup_sim = 0.0
try:
a = float(wup_sim)
if math.isnan(a):
print w1, w2
except:
print w1, w2
matrix_wup[i][j] = wup_sim
matrix_wup[j][i] = wup_sim
j += 1
i += 1
j = 0
similarity_value = np.dot(np.dot(s1_vector, matrix_wup), s2_vector.T) / (
np.linalg.norm(s1_vector) * np.linalg.norm(s2_vector))
#
# print "matrix_wup:\n", matrix_wup
#
# print s1, "#", s2, ":", similarity_value
mylog("1#" + str(similarity_value))
mylog("\n")
mylog("\n")
if __name__ == '__main__':
# for test
# s1 = '''
# dog eat cat
# '''
# s2 = '''
# dog eat cat
# '''
# s1 = CorpusUtil.pre_process_msrp(" ".join(nltk.word_tokenize(s1)))
# s2 = CorpusUtil.pre_process_msrp(" ".join(nltk.word_tokenize(s2)))
i = 0
for line in open(msrp_train_file).readlines():
i += 1
line = line.strip()
split_line = line.split("\t")
flag = int(split_line[0])
s1 = split_line[3].strip()
s2 = split_line[4].strip()
s1 = " ".join(nltk.word_tokenize(s1))
s2 = " ".join(nltk.word_tokenize(s2))
mylog(str(flag) + " ")
similarity(s1, s2)
print i, "行......"
