def cal_greedy(self, sentence1, sentence2, limit_sentence_num):
# 要約文の累積単語数
# word_num = 0
# sentence1のリスト化
list_sentence1 = self.sentence.separate_sentence(sentence1)
# sentence2のリスト化
list_sentence2 = self.sentence.separate_sentence(sentence2)
# 要約文のラベル集合
list_S = []
# 文書集合のラベル集合
list_V = [i for i in range(len(list_sentence1))]
# 文を分かち書きする
list_sentence1_word = [self.sentence.sentence_owakati(sentence)
for sentence in list_sentence1]
# 文を分かち書きする
list_sentence2_word = [self.sentence.sentence_owakati(sentence)
for sentence in list_sentence2]
# リストの文章をベクトルに置き換える
dict_word_to_vector = Filer.readpickle(self.dictpath)
list_sentence1_vector = [self.sentence.sentence_to_vector(sentence_word,
dict_word_to_vector)
for sentence_word in list_sentence1_word]
list_sentence2_vector = [self.sentence.sentence_to_vector(sentence_word,
dict_word_to_vector)
for sentence_word in list_sentence2_word]
# 文間の類似度を計算
arr_similarity = Vector.cal_similarity_matrix(list_sentence1_vector)
# 文書間の類似度を計算
arr_similarity_inv = Vector.cal_mutual_similarity_matrix_inv(list_sentence2_vector, list_sentence1_vector)
# C_V1を計算
arr_C_V1 = self._cal_C_V(arr_similarity)
arr_C_V2 = self._cal_C_V(arr_similarity_inv)
# 文書をクラスタリング
list_sentence_labels = Vector.cal_clustering(list_sentence1_vector,
self.K)
# 劣モジュラ関数の修正貪欲法
while len(list_S) < limit_sentence_num:
# f_doc / c のスコアを記録するための変数
score_tmp = 0
# その時点で最高スコアを記録している文iを記録するための変数
sentence_num_tmp = None
for i in list_V:
# 文iを１つずつ加えてf_docを計算
list_S.append(i)
f_doc = self._cal_f_doc(list_S,
arr_similarity,
arr_similarity_inv,
arr_C_V1,
arr_C_V2,
list_sentence_labels,
self.gamma,
self.epsilon,
self.lamda)
# 計算後に文iを集合Sから除く
list_S.pop()
# f_doc / c がその時点の最高スコアなら記録
if f_doc > score_tmp:
score_tmp = f_doc
sentence_num_tmp = i
# 最高スコアを記録した文iを集合Sに加え、集合Vから外す
list_S.append(sentence_num_tmp)
list_V.remove(sentence_num_tmp)
# list_Sをソート
list_S_rev = sorted(list_S, key=lambda x: x)
# list_S, list_sentence_Sのセット
self.list_sentence_S = [list_sentence1[i] for i in list_S_rev]
self.list_S = list_S
self.list_sentence_V = list_sentence1
def cal_greedy(self, sentence, limit_sentence_num):
# 要約文の累積単語数
# word_num = 0
# sentenceのリスト化
list_sentence = self.sentence.separate_sentence(sentence)
# 要約文のラベル集合
list_S = []
# 文書集合のラベル集合
list_V = [i for i in range(len(list_sentence))]
# 文を分かち書きする
list_sentence_word = [self.sentence.sentence_owakati(sentence)
for sentence in list_sentence]
# リストの文章をベクトルに置き換える
dict_word_to_vector = Filer.readpickle(self.dictpath)
list_sentence_vector = [self.sentence.sentence_to_vector(sentence_word,
dict_word_to_vector)
for sentence_word in list_sentence_word]
# 文間の類似度を計算
arr_similarity = Vector.cal_similarity_matrix(list_sentence_vector)
# C_Vを計算
arr_C_V = self._cal_C_V(arr_similarity)
# 文書をクラスタリング
list_sentence_labels = Vector.cal_clustering(list_sentence_vector,
self.K)
# 劣モジュラ関数の修正貪欲法
# while limit_word_num > word_num:
while len(list_S) < limit_sentence_num:
# f_doc / c のスコアを記録するための変数
score_tmp = 0
# その時点で最高スコアを記録している文iを記録するための変数
sentence_num_tmp = None
for i in list_V:
# 文iを１つずつ加えてf_docを計算
list_S.append(i)
f_doc = self._cal_f_doc(list_S,
arr_similarity,
arr_C_V,
list_sentence_labels,
self.gamma,
self.lamda)
# 計算後に文iを集合Sから除く
list_S.pop()
# f_doc / c がその時点の最高スコアなら記録
"""
if f_doc / len(list_sentence_word[i]) > score_tmp:
score_tmp = f_doc / len(list_sentence_word[i])
sentence_num_tmp = i
"""
if f_doc > score_tmp:
score_tmp = f_doc
sentence_num_tmp = i
# 最高スコアを記録した文iを集合Sに加え、集合Vから外す
list_S.append(sentence_num_tmp)
list_V.remove(sentence_num_tmp)
# 集合Sの累積単語数を更新する
# word_num = sum([len(list_sentence_word[i]) for i in list_S])
# list_Sをソート
list_S_rev = sorted(list_S, key=lambda x: x)
# list_S, list_sentence_S, list_sentence_Vのセット
self.list_sentence_S = [list_sentence[i] for i in list_S_rev]
self.list_S = list_S_rev
self.list_sentence_V = list_sentence
