forked from mattzheng/LangueOne
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtermsRecognition.py
343 lines (298 loc) · 13.6 KB
/
termsRecognition.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
# -*- coding: utf-8 -*-
import sys
#reload(sys)
#sys.setdefaultencoding('utf8')
import warnings
warnings.filterwarnings("ignore")
import regex as re
import numpy as np
import math
import pandas as pd
import jieba
from tqdm import tqdm
class termsRecognition(object):
def __init__(self, content='', topK=-1, tfreq=10, tDOA=0, tDOF=0, is_jieba= False,mode = [1]):
'''
参数：
content: 待成词的文本
maxlen: 词的最大长度
topK: 返回的词数量
tfreq: 频数阈值
tDOA: 聚合度阈值
tDOF: 自由度阈值
mode：词语生成模式，一共四种模式，其中第二种模式比较好,一定要写成[1]
diction:字典，第一批Jieba分词之后的内容
idf_diction:在第一批字典之后，又生成一批tuple words 的idf，计算方式是，两个词语的平均
punct:标点符号，Jieba分词之后删除
步骤：
jieba_tuples_generator， 利用Jieba分词，并去除标点符号，去除清除''(写入self.jieba_content)，利用wordsGenerator函数生成词语对（四种模式）(写入self.tuple_content)
word_get_frequency_idf，计算freq 以及贴idf,同时生成'left'/right框，把词语对（self.tuple_content)）写入result(★ 主要写入部分)
get_doa：只输入result,计算数据的doa，直接更新result中的['doa']
word_get_dof:只输入result,计算数据的dof，直接更新result中的['dof'],左熵的文字,右熵的文字
get_score,只输入result,更新result中的['scores']
可用的函数:
get_idf,文档的IDF计算
wordsGenerator,生成词语对
get_entropy计算左、右熵值，填充result
'''
self.content = content
self.jieba_content = ''
self.tuple_content = [] # 多项式文本
self.topK = topK
self.tfreq = tfreq
self.tDOA = tDOA
self.tDOF = tDOF
self.mode = mode
self.is_jieba = is_jieba
self.diction = pd.DataFrame()
self.idf_diction = {}
self.punct = ''':!),.:;?]}¢'"、。〉》」』】〕〗〞︰︱︳﹐､﹒﹔﹕﹖﹗﹚﹜﹞！），．：；？｜｝︴︶︸︺︼︾﹀﹂﹄﹏､～￠々‖•·ˇˉ―--′’”([{£¥'"‵〈《「『【〔〖（［｛￡￥〝︵︷︹︻︽︿﹁﹃﹙﹛﹝（｛“‘-—_…'''
self.result = {}
def jieba_tuples_generator(self):
print('Cut the words ...')
if self.is_jieba:
self.diction = self.get_idf(self.content) # 只有多行的才能计算IDF
if isinstance (self.content,tuple):
self.jieba_content = sum(self.content,())
elif isinstance (self.content,list):
self.jieba_content = sum(self.content,[])
else:
if isinstance (self.content,str):
self.jieba_content = list(jieba.cut(self.content))
self.jieba_content = list(map( lambda s: ''.join(filter(lambda x: x not in self.punct, s)) , self.jieba_content ))
self.jieba_content = list(pd.Series(self.jieba_content) [ [i !='' for i in self.jieba_content] ].values) # 清除''
else:
self.jieba_content = [list(jieba.cut(i)) for i in self.content]
self.diction = self.get_idf(self.jieba_content) # 只有多行的才能计算IDF
self.jieba_content = list(map( lambda s: ''.join(filter(lambda x: x not in self.punct, s)) , sum(self.jieba_content,[]) ))
self.jieba_content = list(pd.Series(self.jieba_content) [ [i !='' for i in self.jieba_content] ].values) # 清除''
print('Calculate tuple words ...')
for t_mode in self.mode:
self.tuple_content.extend( self.wordsGenerator( self.jieba_content,t_mode) )
def word_get_frequency_idf(self):
# 作用：计算freq 以及贴idf,同时生成'left'/right框，直接写入result
# 词粒度
print ('Calculate Frequency for each possible words')
reg = [i[1] for i in self.tuple_content] #
for r in reg:
if r in self.result:
self.result[r]['freq'] += 1
self.result[r]['idf'] = self.idf_diction[r]
else:
self.result[r] = {'left':[], 'right':[]}
self.result[r]['freq'] = 1
self.result[r]['idf'] = self.idf_diction[r]
def docs(self,w, D):
'''
w,词;
D,整个文档，分好词的，譬如[('你','好'),('我','们')....]
计算含有w词的文档数量
'''
c = 0
for d in D:
if w in d:
c = c + 1;
return c
def get_idf(self,jieba_content):
'''
作用：生成、计算idf
参考：http://www.voidcn.com/article/p-mhebqvic-qq.html
输入：分行的文本内容;
输出：dataframe,分别有:words / tf / idf 三列
'''
print('Calculate IDF ...')
try:
diction = pd.DataFrame( list(set(sum(jieba_content,[])) ) , columns = ['words'])
except:
diction = pd.DataFrame( list(set(sum(jieba_content,())) ) , columns = ['words'])
diction['tf'] = list(map(lambda x : self.docs(x,jieba_content) , diction['words'] ))
n = len(jieba_content)
diction['idf'] = [math.log(i) for i in n*1.0 / (diction['tf'] + 1 ) ]
return diction
def get_doa(self, base=2):
'''
pa = 当前词/ 文档数量
pl = 前词
pr = 后词
P(S)/(P(sl)×P(sr))的最小值，取对数之后即可作为聚合度的衡量
'''
# 使用信息熵衡量每个词语的聚合度
print ('Calculate DOA for each possible words')
for key, value in tqdm(self.result.items()):
if len(key) == 1:
self.result[key]['doa'] = 0
continue
doa = 99999
for x in range(1, len(key)):
try:
pa = float(self.result[key]['freq']) / len(self.content)
pl = float(self.result[key[:x]]['freq']) / len(self.content)
pr = float(self.result[key[x:]]['freq']) / len(self.content)
td = math.log(pa / (pl * pr), base)
if td < doa:
doa = td
except:
pass
else:
pass
finally:
pass
self.result[key]['doa'] = doa
def wordsGenerator(self,base_text , mode = 1):
'''
输入 ：jieba.cut分词
输出 ：List,[('f', '司法', '解'),('司', '法解', '释')]
譬如：相对来说,1比较适合
'这个函数真的比较特殊'-> '这个,函数,真的,比较,特殊'
mode == 1:
词,词+词，词 -->'这个,函数真的,比较' + '函数,真的比较,特殊'
mode == 2:（效果最佳）
词，词+词+词，词 -->'这个,函数真的比较,特殊'
mode == 3:
词，字+词，词 / 词，词+ 字，词 -->'这个,个函数,真的' , '这个,函数真,真的'
mode == 22(fail):
字+字，词+词，字+字 -->'这个,函数真的,比较' + '函数,真的比较,特殊'
mode == 4:
词，词+词+词+词，词 -->'这个,函数真的比较特殊'
'''
base_word = []
if mode == 1 :
for i in tqdm(range(1,len(base_text)-3)):
base_word.append((base_text[i-1] , base_text[i] + ' ' + base_text[i+1] , base_text[i+2]))
# 计算idf
#print('word generator mode 1 ,Round at %s ...' %i)
self.idf_diction[base_text[i] + ' ' + base_text[i+1]] = np.hstack((self.diction[ (self.diction['words']==base_text[i]) ]['idf'].values,\
self.diction[ (self.diction['words']==base_text[i+1]) ]['idf'].values)).mean()
if mode == 2 :
for i in tqdm(range(1,len(base_text)-4)):
base_word.append((base_text[i-1] , base_text[i] + ' ' + base_text[i+1] + ' ' + base_text[i+2] , base_text[i+3]))
# 计算idf
#print('word generator mode 2 ,Round %s ... ' %i)
self.idf_diction[ base_text[i] + ' ' + base_text[i+1] + ' ' + base_text[i+2]] = np.hstack((self.diction[ (self.diction['words']==base_text[i]) ]['idf'].values,\
self.diction[ (self.diction['words']==base_text[i+1]) ]['idf'].values,\
self.diction[ (self.diction['words']==base_text[i+2]) ]['idf'].values)).mean()
if mode == 3:
for i in tqdm(range(1,len(base_text)-3)):
base_word.append(( base_text[i-1] , base_text[i-1][-1] + ' ' + base_text[i] , base_text[i+1] ))
base_word.append(( base_text[i-1] , base_text[i] + ' ' + base_text[i+1][0], base_text[i+1] ))
# 计算idf
#print('word generator mode 3 ,Round at %s ...' %i)
self.idf_diction[base_text[i-1][-1] + ' ' + base_text[i]] = np.hstack((self.diction[ (self.diction['words']==base_text[i]) ]['idf'].values,\
self.diction[ (self.diction['words']==base_text[i-1]) ]['idf'].values)).mean()
self.idf_diction[base_text[i] + ' ' + base_text[i+1][0]] = np.hstack((self.diction[ (self.diction['words']==base_text[i]) ]['idf'].values,\
self.diction[ (self.diction['words']==base_text[i+1]) ]['idf'].values)).mean()
if mode == 4 :
for i in tqdm(range(1,len(base_text)-5)):
base_word.append((base_text[i-1] , base_text[i] + ' ' + base_text[i+1] + ' ' + base_text[i+2] + ' ' + base_text[i+3], base_text[i+4]))
# 计算idf
#print('word generator mode 4 ,Round at %s ...' %i)
self.idf_diction[base_text[i] + ' ' + base_text[i+1] + ' ' + base_text[i+2] + ' ' + base_text[i+3]] = \
np.hstack((self.diction[ (self.diction['words']==base_text[i]) ]['idf'].values,\
self.diction[ (self.diction['words']==base_text[i+1]) ]['idf'].values,\
self.diction[ (self.diction['words']==base_text[i+2]) ]['idf'].values,\
self.diction[ (self.diction['words']==base_text[i+3]) ]['idf'].values)).mean()
return base_word
def word_get_dof(self):
'''
根据tuple_content的词语对，填充result中的['left'] / ['right']两列
每个词，根据填充好的left/ right的词语，通过get_entropy计算左、右熵值，填充result
'''
# 使用信息熵衡量每个词语的自由度
print('Calculate DOF for each possible words')
reg = self.tuple_content
for r in reg:
self.result[r[1]]['left'].append(r[0])
self.result[r[1]]['right'].append(r[2])
for key, value in tqdm(self.result.items()):
left = self.get_entropy(self.delete_boxex(value['left']))
right = self.get_entropy(self.delete_boxex(value['right']))
if left < right:
self.result[key]['dof'] = left
else:
self.result[key]['dof'] = right
def delete_boxex(self ,strings):
[i != [] for i in strings]
return list(pd.Series(strings) [ [i != [] for i in strings ] ] .values)
def get_entropy(self, data, base=2):
'''
根据tuple_content的词语对，填充result中的['left'] / ['right']两列
每个词，根据填充好的left/ right的词语，通过get_entropy计算左、右熵值，填充result
'''
tmp = {}
for item in data:
if not item in tmp:
tmp[item] = 1.0
else:
tmp[item] += 1.0
for key, value in tmp.items():
tmp[key] /= float(len(data))
result = 0.0
for key, value in tmp.items():
result += value * math.log(value, base)
if result < 0:
result = -result
return result
def get_score(self):
# 将频数、聚合度、自由度归一化，并计算总得分
print ('Calculate Score for each possible words')
for key, value in tqdm(self.result.items()):
if value['freq'] <= self.tfreq or value['doa'] <= self.tDOA or value['dof'] <= self.tDOF:
self.result[key]['score'] = 0
else:
self.result[key]['score'] = value['freq'] * value['doa'] * value['dof'] * value['idf']
def generate_word(self):
'''
步骤：
jieba_tuples_generator， 利用Jieba分词，并去除标点符号，去除清除''(写入self.jieba_content)，利用wordsGenerator函数生成词语对（四种模式）(写入self.tuple_content)
word_get_frequency_idf，计算freq 以及贴idf,同时生成'left'/right框，把词语对（self.tuple_content)）写入result(★ 主要写入部分)
get_doa：只输入result,计算数据的doa，直接更新result中的['doa']
word_get_dof:只输入result,计算数据的dof，直接更新result中的['dof'],左熵的文字,右熵的文字
get_score,只输入result,更新result中的['scores']
'''
self.jieba_tuples_generator()
self.word_get_frequency_idf()
self.get_doa()
self.word_get_dof()
self.get_score()
result = sorted(self.result.items(), key=lambda d:d[1]['score'], reverse=True)
if self.topK == -1:
return result
else:
return result[:self.topK]
def get_result(self):
result = []
for key,values in self.result.items():
result.append([key,values['dof'],values['doa'], values['freq'] , values['score'] ,values['idf'] ])
return pd.DataFrame(result,columns = ['key','dof','doa','freq','score','idf'])
def part_found(self):
'''
部分发现:
只计算词频freq以及idf值
计算之后，生成dataframe表格
'''
print('local found ... ')
generator.jieba_tuples_generator()
generator.word_get_frequency_idf()
result_dataframe = pd.DataFrame(generator.result).T
result_dataframe['words'] = result_dataframe.index
result_dataframe.reset_index(inplace=True,drop=True)
return result_dataframe
# --------------------------- 词粒度 评估 ---------------------------
if __name__ == '__main__':
data = pd.read_csv('toutiao_data.csv',encoding = 'utf-8')
# ----------- 未分词 -----------
generator = termsRecognition(content = data['new_title'][:10000] ,is_jieba=False, topK = 20 , mode = [1]) # 文字版
# 全部发现
result_dict = generator.generate_word()
result_dataframe = generator.get_result()
# 部分发现
result_dataframe = generator.part_found()
# ----------- 已分词 -----------
def not_nan(obj):
return obj == obj
keywords = []
for word in tqdm(data.new_keyword):
if not_nan(word):
keywords.append(word.split(','))
generator = termsRecognition(content = keywords[:1000] , is_jieba=True , topK = 20,mode = [1]) #图像版
result_dataframe = generator.part_found()