def run_experiment_with_rake():
print "\nBegin experiment using RAKE algorithm..."
# RAKE: predict keyword dengan RAKE, ambil words dengan RAKE skor tertinggi
rake = RakeKeywordExtractor()
tweets_rake['keyword'] = tweets_rake.apply(lambda t: rake.extract_keyword(
rake.extract_candidates(t['text'], incl_scores=True)),
axis=1)
# RAKE: infer aspect dengan aspect mapping, dengan similarity terbesar
tweets_rake['selected_keyword'] = tweets_rake.apply(
lambda t: asp.find_nearest_inferred_aspect(t['keyword'], emb)[1],
axis=1)
tweets_rake['inferred_aspect'] = tweets_rake.apply(
lambda t: asp.find_nearest_inferred_aspect(t['keyword'], emb)[0],
axis=1)
tweets_rake['gold_aspect'] = tweets_rake.apply(
lambda t: asp.INVERTED_ASPECTS[t['inferred_aspect']], axis=1)
tweets_rake.to_csv('dump/result_rake.csv', encoding='utf-8', index=False)
# RAKE: Evaluasi dengan accuracy
eva_rake = Evaluation(tweets_rake)
conf_matrix = eva_rake.build_confusion_matrix(tweets_rake)
print "Confusion matrix:"
print conf_matrix
print "Accuracy using RAKE algorithm: {}".format(eva_rake.accuracy())
print "Average Precision using RAKE algorithm: {}".format(
eva_rake.average_precision())
print "Average Recall using RAKE algorithm: {}".format(
eva_rake.average_recall())
def run_experiment_with_tfidf(tweets_tfidf):
print "\nBegin experiment using TF-IDF weighting algorithm..."
# TF-IDF: cari keyword dengan TF-IDF, ambil yang single word aja dengan bobot tertinggi
tfidf = TfidfKeywordExtractor()
tfidf_weight = tfidf.fit_transform(tweets_tfidf)
tfidf_weight['keyword'] = tfidf_weight.idxmax(axis=1)
# MUST BE after extracting keyword
# OTHERWISE, the keyword will be "tweet_no" for all tweets
tfidf_weight = tfidf_weight.reset_index().rename(
columns={'index': 'tweet_no'})
tfidf_weight['tweet_no'] = tfidf_weight['tweet_no'] + 1
tfidf_weight = tfidf_weight[['tweet_no', 'keyword']]
tfidf_weight.to_csv('tfidf_keyword.csv', encoding='utf-8', index=False)
tweets_tfidf = tweets_tfidf.reset_index().rename(
columns={'index': 'tweet_no'})
tweets_tfidf['tweet_no'] = tweets_tfidf['tweet_no'] + 1
tweets_tfidf.to_csv('tweets_tfidf.csv', encoding='utf-8', index=False)
tweets_tfidf = pd.merge(tweets_tfidf,
tfidf_weight,
how='left',
on='tweet_no')
tweets_tfidf.to_csv('tweets_tfidf_after_merge.csv',
encoding='utf-8',
index=False)
# TF-IDF: infer aspect dengan aspect mapping, dengan similarity terbesar
tweets_tfidf['selected_keyword'] = tweets_tfidf.apply(
lambda t: asp.find_nearest_inferred_aspect(t['keyword'], emb)[1],
axis=1)
tweets_tfidf['inferred_aspect'] = tweets_tfidf.apply(
lambda t: asp.find_nearest_inferred_aspect(t['keyword'], emb)[0],
axis=1)
tweets_tfidf['gold_aspect'] = tweets_tfidf.apply(
lambda t: asp.INVERTED_ASPECTS[t['inferred_aspect']], axis=1)
tweets_tfidf.to_csv('dump/result_tfidf.csv', encoding='utf-8', index=False)
# RAKE: Evaluasi dengan accuracy
eva_tfidf = Evaluation(tweets_tfidf)
conf_matrix = eva_tfidf.build_confusion_matrix(tweets_tfidf)
print "Confusion matrix:"
print conf_matrix
print "Accuracy using TF-IDF weighting algorithm: {}".format(
eva_tfidf.accuracy())
print "Average Precision using TF-IDF weighting algorithm: {}".format(
eva_tfidf.average_precision())
print "Average Recall using TF-IDF weighting algorithm: {}".format(
eva_tfidf.average_recall())
def calculate_score_and_show_result(self, w2v_150_socres, w2v_300_scores,
bert_768_scores):
e = Evaluation()
w2v_precision_150 = e.average_precision(w2v_150_socres, 2)
w2v_ndcg_150 = e.average_ndcg(w2v_150_socres)
w2v_precision_300 = e.average_precision(w2v_300_scores, 2)
w2v_ndcg_300 = e.average_ndcg(w2v_300_scores)
bert_precision_768 = e.average_precision(bert_768_scores, 2)
bert_ndcg_768 = e.average_ndcg(bert_768_scores)
print(w2v_precision_150)
print(w2v_ndcg_150)
print(w2v_precision_300)
print(w2v_ndcg_300)
print(bert_precision_768)
print(bert_ndcg_768)
plt.title("Semantic Prediction in Frequent Emotions and Events")
plt.xlabel(
"The number of predict semantic for each emotions or events")
plt.ylabel("Average Precision")
plt.plot(range(1, 11),
w2v_precision_150,
"-o",
color='r',
label="e2v-w2v-sg 150 dimension") # 畫出平均數值
plt.plot(range(1, 11),
w2v_precision_300,
"-o",
color='b',
label="e2v-w2v-sg 300 dimension") # 畫出平均數值
plt.plot(range(1, 11),
bert_precision_768,
"-o",
color='g',
label="e2v-bert 768 dimension") # 畫出平均數值
plt.legend(loc="best")
# save image
plt.savefig('image/precision.png')
plt.close()
plt.title("Semantic Prediction in Frequent Emotions and Events")
plt.xlabel(
"The number of predict semantic for each emotions or events")
plt.ylabel("Average NDCG")
plt.plot(range(1, 11),
w2v_ndcg_150,
"-o",
color='r',
label="e2v-w2v-sg 150 dimension") # 畫出平均數值
plt.plot(range(1, 11),
w2v_ndcg_300,
"-o",
color='b',
label="e2v-w2v-sg 300 dimension") # 畫出平均數值
plt.plot(range(1, 11),
bert_ndcg_768,
"-o",
color='g',
label="e2v-bert 768 dimension") # 畫出平均數值
plt.legend(loc="best")
# save image
plt.savefig('image/ndcg.png')
plt.close()
