def title_cluster(df, save_df=False):
"""按新闻标题聚类"""
df_title = df.copy()
df_title = title_preprocess(df_title)
word_library_list = counter.get_word_library(df_title['title_cut'])
single_frequency_words_list = counter.get_single_frequency_words(df_title['title_cut'])
max_features = len(word_library_list) - len(single_frequency_words_list) // 2
title_matrix = modeling.feature_extraction(df_title['title_'], vectorizer='CountVectorizer',
vec_args={'max_df': 1.0, 'min_df': 1, 'max_features': max_features})
title_dbscan = modeling.get_cluster(title_matrix, cluster='DBSCAN',
cluster_args={'eps': 0.4, 'min_samples': 4, 'metric': 'cosine'})
title_labels = modeling.get_labels(title_dbscan)
df_title['title_label'] = title_labels
df_non_outliers = modeling.get_non_outliers_data(df_title, label_column='title_label')
title_label_num = counter.get_num_of_value_no_repeat(df_non_outliers['title_label'].tolist())
print('按新闻标题聚类，一共有%d个簇(不包括离群点)' % title_label_num)
title_rank = modeling.label2rank(title_labels)
df_title['title_rank'] = title_rank
for i in range(1, title_label_num + 1):
df_ = df_title[df_title['title_rank'] == i]
title_top_list = counter.get_most_common_words(df_['title_cut'], top_n=10)
print(title_top_list)
if save_df:
df_title.drop(['content', 'title_', 'title_label'], axis=1, inplace=True)
news_crawler.save_news(df_title, os.path.join(results_path, 'df_title_rank.csv'))
return df_title
def get_key_words():
df_title = news_crawler.load_news(
os.path.join(results_path, 'df_title_rank.csv'))
df_content = news_crawler.load_news(
os.path.join(results_path, 'df_content_rank.csv'))
df_title['title_cut'] = df_title['title_cut'].map(eval)
df_content['content_cut'] = df_content['content_cut'].map(eval)
df_title_content = df_title.copy()
df_title_content['content_cut'] = df_content['content_cut']
df_title_content['content_rank'] = df_content['content_rank']
df_title_content = modeling.get_non_outliers_data(
df_title_content, label_column='title_rank')
title_rank_num = counter.get_num_of_value_no_repeat(
(df_title_content['title_rank']))
for i in range(1, title_rank_num + 1):
df_i = df_title_content[df_title_content['title_rank'] == i]
title = '\n'.join(df_i['title'].tolist())
title = modeling.get_key_sentences(title, num=1)
print('热点：', title)
content_rank = [k for k in df_i['content_rank']]
content_rank = set(content_rank)
for j in content_rank:
df_j = df_i[df_i['content_rank'] == j]
most_commmon_words = counter.get_most_common_words(
df_j['content_cut'], top_n=20, min_frequency=5)
if len(most_commmon_words) > 0:
print('相关词汇：', most_commmon_words)
def content_cluster(df, df_save=False):
"""按新闻内容聚类"""
df_content = df.copy()
df_content = content_preprocess(df_content)
word_library_list = counter.get_word_library(df_content['content_cut'])
single_frequency_words_list = counter.get_single_frequency_words(df_content['content_cut'])
max_features = len(word_library_list) - len(single_frequency_words_list) // 2
content_matrix = modeling.feature_extraction(df_content['content_'], vectorizer='CountVectorizer',
vec_args={'max_df': 0.95, 'min_df': 1, 'max_features': max_features})
content_dbscan = modeling.get_cluster(content_matrix, cluster='DBSCAN',
cluster_args={'eps': 0.35, 'min_samples': 4, 'metric': 'cosine'})
content_labels = modeling.get_labels(content_dbscan)
df_content['content_label'] = content_labels
df_non_outliers = modeling.get_non_outliers_data(df_content, label_column='content_label')
content_label_num = counter.get_num_of_value_no_repeat(df_non_outliers['content_label'].tolist())
print('按新闻内容聚类，一共有%d个簇(不包括离群点)' % content_label_num)
content_rank = modeling.label2rank(content_labels)
df_content['content_rank'] = content_rank
for i in range(1, content_label_num + 1):
df_ = df_content[df_content['content_rank'] == i]
content_top_list = counter.get_most_common_words(df_['content_cut'], top_n=15, min_frequency=1)
print(content_top_list)
if df_save:
df_content.drop(['content_', 'content_label'], axis=1, inplace=True)
news_crawler.save_news(df_content, os.path.join(results_path, 'df_content_rank.csv'))
return df_content
def get_wordcloud(df, rank_column, text_list_column):
"""
按照不同的簇生成每个簇的词云
:param df: pd.DataFrame，带有排名和分词后的文本列表数据
:param rank_column: 排名列名
:param text_list_column: 分词后的文本列表列名
"""
df_non_outliers = modeling.get_non_outliers_data(df, label_column=rank_column)
label_num = counter.get_num_of_value_no_repeat(df_non_outliers[rank_column].tolist())
for i in range(1, label_num + 1):
df_ = df[df[rank_column] == i]
list_ = counter.flat(df_[text_list_column].tolist())
modeling.list2wordcloud(list_, save_path=os.path.join(results_path, rank_column, '%d.png' % i),
font_path=os.path.join(fonts_path, 'yw.ttf'))
