def calculate_pearson_for_sts131(stemming, lowercase, stopwords,
remove_notalpha, analyze_measure):
stss131_df = pd.read_csv("./data/stss131.csv", sep=",")
# First we run through the dataset row by row, and append the results to a list
list_of_similarity_values = []
for s1, s2 in zip(stss131_df.Sentence1, stss131_df.Sentence2):
if analyze_measure == "original":
similarity_score = wni.calculate_path_similarity_for_sentences(
sentence1=s1,
sentence2=s2,
stemming=stemming,
lowercase=lowercase,
stopwords=stopwords,
remove_notalpha=remove_notalpha)
print("Similarity score of the current sentence pair is:",
similarity_score)
elif analyze_measure == "hyp-ed method":
similarity_score = csm.count_custom_similarity_measure(
sentence1=s1,
sentence2=s2,
stemming=stemming,
lowercase=lowercase,
stopwords=stopwords,
remove_notalpha=remove_notalpha)
print("Similarity score of the current sentence pair is:",
similarity_score)
elif analyze_measure == "semantic text similarity method":
sentence1_words = preprocess(s1, stemming, lowercase, stopwords,
remove_notalpha)
sentence2_words = preprocess(s2, stemming, lowercase, stopwords,
remove_notalpha)
s1_word_list, s2_word_list = [], []
for word1 in sentence1_words:
s1_word_list.append(word1[0])
for word2 in sentence2_words:
s2_word_list.append(word2[0])
similarity_score = soc.soc_pmi(s1_word_list, s2_word_list)
print("Similarity score of the current sentence pair is:",
similarity_score)
list_of_similarity_values.append(similarity_score)
stss131_df["R"] = pd.Series(list_of_similarity_values)
pearson_corr = pearsonr(stss131_df.X, stss131_df.R)
print("Pearson correlation between Rater mean: {corr}, p-value {pval}".
format(corr=pearson_corr[0], pval=pearson_corr[1]))
def calculate_path_similarity_for_sentences(sentence1, sentence2, stemming,
lowercase, stopwords,
remove_notalpha):
sentence1_words = preprocess(sentence=sentence1,
use_stemmer=stemming,
use_lowercase=lowercase,
use_stopwords=stopwords,
remove_nonalpha=remove_notalpha)
sentence2_words = preprocess(sentence=sentence2,
use_stemmer=stemming,
use_lowercase=lowercase,
use_stopwords=stopwords,
remove_nonalpha=remove_notalpha)
sentence1_score = 0
sentence2_score = 0
for word1 in sentence1_words:
max_score = 0
word1_synsets = retrieve_synset_list_for_word(word1[0])
for word2 in sentence2_words:
word2_synsets = retrieve_synset_list_for_word(word2[0])
score = calculate_path_similarity_for_synset_lists(
word1_synsets, word2_synsets)
if score > max_score:
max_score = score
sentence1_score += max_score
for word2 in sentence2_words:
max_score = 0
word2_synsets = retrieve_synset_list_for_word(word2[0])
for word1 in sentence1_words:
word1_synsets = retrieve_synset_list_for_word(word1[0])
score = calculate_path_similarity_for_synset_lists(
word2_synsets, word1_synsets)
if score > max_score:
max_score = score
sentence2_score += max_score
# Count the means for each sentence, add them together and divide by number of sentences
sentence1_score_mean = sentence1_score / len(sentence1_words)
sentence2_score_mean = sentence2_score / len(sentence2_words)
sentence_similarity_score = (sentence1_score_mean +
sentence2_score_mean) / 2
return round(sentence_similarity_score, 2)
def test_preprocessing_remove_stopwords():
test_sentence = "This is A TEST SENtence, to Make Sure Lower works"
result_sentence = ["TEST", "SENtence", "Make", "Sure", "Lower", "works"]
preprocessed_sentence = swp.preprocess(sentence=test_sentence,
use_stemmer=False,
use_lowercase=False,
use_stopwords=True,
remove_nonalpha=False)
preprocessed_sentence_just_words = [
word[0] for word in preprocessed_sentence
]
assert result_sentence == preprocessed_sentence_just_words
def test_preprocessing_tolower_remove_nonalpha():
test_sentence = "This is A TEST SENtence, to Make Sure Lower works"
result_sentence = [
"this", "is", "a", "test", "sentence", "to", "make", "sure", "lower",
"works"
]
preprocessed_sentence = swp.preprocess(sentence=test_sentence,
use_stemmer=False,
use_lowercase=True,
use_stopwords=False,
remove_nonalpha=True)
preprocessed_sentence_just_words = [
word[0] for word in preprocessed_sentence
]
assert result_sentence == preprocessed_sentence_just_words
stemming=args.stem,
lowercase=args.lowercase,
stopwords=args.stopwords,
remove_notalpha=args.nonalpha)
print("Sentence similarities: " + str(sentence_similarity_score))
elif args.measure == "hyp-ed method":
logging.info("Calculating path similarity for sentences using hypernyms and hyponyms.")
sentence_similarity_score = csm.count_custom_similarity_measure(sentence1=args.sentence1,
sentence2=args.sentence2,
stemming=args.stem, lowercase=args.lowercase,
stopwords=args.stopwords,
remove_notalpha=args.nonalpha)
print("Sentence similarities: " + str(sentence_similarity_score))
elif args.measure == "semantic text similarity method":
logging.info("Calculating path similarity for sentences using SOC PMI Alogrithm.")
sentence1_words = preprocess(sentence=args.sentence1, use_stemmer=args.stem, use_lowercase=args.lowercase,
use_stopwords=args.stopwords, remove_nonalpha=args.nonalpha)
sentence2_words = preprocess(sentence=args.sentence2, use_stemmer=args.stem, use_lowercase=args.lowercase,
use_stopwords=args.stopwords, remove_nonalpha=args.nonalpha)
s1_word_list, s2_word_list = [], []
for word1 in sentence1_words:
s1_word_list.append(word1[0])
for word2 in sentence2_words:
s2_word_list.append(word2[0])
sentence_similarity_score = soc.soc_pmi(s1_word_list, s2_word_list)
print("Sentence similarities: " + str(sentence_similarity_score))