def generate_features(stances, dataset, name, bow_vectorizer, tfreq_vectorizer,
tfidf_vectorizer):
h, b, y = [], [], []
for stance in stances:
y.append(LABELS.index(stance['Stance']))
h.append(stance['Headline'])
b.append(dataset.articles[stance['Body ID']])
X_overlap = gen_or_load_feats(word_overlap_features, h, b,
"features/overlap." + name + ".npy")
# X_refuting_body: I add a new refuting feature about the existence of refuting words in the body
X_refuting_head, X_refuting_body = refuting_features(h, b)
X_polarity = gen_or_load_feats(polarity_features, h, b,
"features/polarity." + name + ".npy")
X_hand = gen_or_load_feats(hand_features, h, b,
"features/hand." + name + ".npy")
# X_senti_head: The sentiment vector of the headline;
# X_senti_body: The sentiment vector of the body;
# X_senti_cos : The cosine similarity between the sentiment vectors of the headline and body.
X_senti_head, X_senti_body, X_senti_cos = sentiment_features(h, b)
# X_tf_cos : The cosine similarity between the TF vectors of the headline and body
# X_tf_idf_cos : The cosine similarity between the TF-IDF vectors of the headline and body.
X_tf_cos, X_tf_idf_cos = gen_tf_idf_feats(stances, dataset.articles,
bow_vectorizer, tfreq_vectorizer,
tfidf_vectorizer)
X = np.c_[X_hand, X_polarity, X_refuting_head, X_overlap, X_tf_cos]
return X, y
def classify(headline, body, clf):
h = [headline]
b = [body]
X_overlap = word_overlap_features(h, b)
X_refuting = refuting_features(h, b)
X_polarity = polarity_features(h, b)
X_hand = hand_features(h, b)
X = np.c_[X_hand, X_polarity, X_refuting, X_overlap]
predicted = [LABELS[int(a)] for a in clf.predict(X)]
return predicted
def get_feature(self):
dataset = DataSet(path=self.path, name=self.name)
h, b = [], []
stances = dataset.stances
for stance in stances:
h.append(stance['Headline'])
b.append(dataset.articles[stance['Body ID']])
X_overlap = word_overlap_features(h, b)
X_refuting = refuting_features(h, b)
X_polarity = polarity_features(h, b)
X_hand = hand_features(h, b)
X = np.c_[X_hand, X_polarity, X_refuting, X_overlap]
return X
