def dist_tuple(nc_rep1, nc_rep2):
# domain nouns related to the noun compound
# space is a broad measure of similarity
domain_full1 = nc_rep1[0]
domain_full2 = nc_rep2[0]
dist0 = cosine_dist(domain_full1, domain_full2)
# domain nouns related to the head
# space is analagous to the head
domain_head1 = nc_rep1[1]
domain_head2 = nc_rep2[1]
dist1 = cosine_dist(domain_head1, domain_head2)
# verbs related to the noun compound
# space is measuring the action of the modifier on the head
action1 = nc_rep1[2]
action2 = nc_rep2[2]
dist2 = cosine_dist(action1, action2)
# adjectives that act on the head
# space is analagous to modifier
qualifier_head1 = nc_rep1[3]
qualifier_head2 = nc_rep2[3]
dist3 = cosine_dist(qualifier_head1, qualifier_head2)
# adjectives the modifier might be like
# space is analagous to modifier
qualifier_mod1 = nc_rep1[4]
qualifier_mod2 = nc_rep2[4]
dist4 = cosine_dist(qualifier_mod1, qualifier_mod2)
return (dist0, dist1, dist2, dist3, dist4)
def rank_mean_word_vector(X, k):
''' Rank responses based on cosine similarity between mean word vectors '''
n = X.shape[0]
mat = np.zeros((n, n))
mean_q = X['t_q'].apply(lambda x: mean_word_vector(' '.join(x)))
mean_r = X['t_r'].apply(lambda x: mean_word_vector(' '.join(x)))
for (iq, ir) in product(range(n), range(n)):
mat[iq, ir] = cosine_dist(mean_q.values[iq].reshape(1, -1),
mean_r.values[ir].reshape(1, -1))
r = dist_matrix_to_rank_k(mat, k)
return r
def rank_cos_tfifd_vector(X, k):
''' Rank responses based on cosine similarity between tf-ifd vectors '''
n = X.shape[0]
mat = np.zeros((n, n))
vectorizer = TfidfVectorizer()
resp = vectorizer.fit_transform(X['t_r'].apply(lambda x: ' '.join(x)))
ques = vectorizer.transform(X['t_q'].apply(lambda x: ' '.join(x)))
for (iq, ir) in product(range(n), range(n)):
mat[iq, ir] = cosine_dist(ques[iq, :], resp[ir, :])
r = dist_matrix_to_rank_k(mat, k)
return r
def sim_dilation(nc1, nc2):
v1 = nc_dils_dict[nc1]
v2 = nc_dils_dict[nc2]
return cosine_dist(v1, v2)
def sim_sum(nc1, nc2):
v1 = nc_sums_dict[nc1]
v2 = nc_sums_dict[nc2]
return cosine_dist(v1, v2)