-
Notifications
You must be signed in to change notification settings - Fork 0
/
lib_metric.py
433 lines (375 loc) · 15.4 KB
/
lib_metric.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
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
import math
from config import *
def get_dirichlet_prob(tf_t_d, len_d, tf_t_C, len_C, mu):
"""
Computes Dirichlet-smoothed probability
P(t|theta_d) = [tf(t, d) + mu P(t|C)] / [|d| + mu]
:param tf_t_d: tf(t,d)
:param len_d: |d|
:param tf_t_C: tf(t,C)
:param len_C: |C| = \sum_{d \in C} |d|
:param mu: \mu
:return:
"""
if mu == 0: # i.e. field does not have any content in the collection
return 0
else:
p_t_C = tf_t_C / len_C if len_C > 0.0 else 0.0
return (tf_t_d + mu * p_t_C) / (len_d + mu)
def lmSim(lt_obj1,entityObj,field,w2vmodel,mode,lucene_obj):
# subquery x et[0..n-1]
totalSim=0.0
if mode=='mul':
totalSim=1.0
term_freq=entityObj.term_freq
len_C_f = lucene_obj.get_coll_length(field)
mu=lucene_obj.get_avg_len(field)
cnt=0
# iterate each t in term_freq and compare similarity
for i in range(lt_obj1.length):
qt=lt_obj1.term[i]
# compute p(t|De)
localSim=term_freq.get(qt,0)
if localSim>0.0:
cnt+=1
len_d_f = entityObj.length
tf_t_d_f = localSim
tf_t_C_f = lucene_obj.get_coll_termfreq(qt, field)
p_t_d=get_dirichlet_prob(tf_t_d_f, float(len_d_f), float(tf_t_C_f), float(len_C_f), mu)
# compute f(p(t1|De),p(t2|De)...)
if mode=='sum':
totalSim+=p_t_d
elif mode=='max':
totalSim=max(p_t_d,totalSim)
elif mode=='mul':
if p_t_d>0.0:
totalSim*=p_t_d
elif mode=='log_sum' or mode=='log_sum_avg':
if p_t_d>0.0:
totalSim+=math.log(p_t_d)
if mode=='log_sum_avg' and cnt>0:
totalSim/=float(cnt)
return totalSim
def mlmSim(lt_obj1,entityObj,lucene_obj):
# need every field representation instead of single lt_obj for entity
# subquery x et[0..n-1]
len_C_f={}
mu={}
mlm_weights={}
for f in LIST_F:
len_C_f[f]=lucene_obj.get_coll_length(f)
mu[f]=lucene_obj.get_avg_len(f)
mlm_weights[f]=1.0/len(LIST_F)
if MODEL_NAME=='mlm-tc':
mlm_weights={'stemmed_names':0.2,'stemmed_catchall':0.8} if USED_QUERY_VERSION=='stemmed_raw_query' else {'names':0.2,'catchall':0.8}
totalSim=0.0
for i in range(lt_obj1.length):
qt=lt_obj1.term[i]
localSim=0.0
# compute p(t|Df)
for f in LIST_F:
len_d_f = entityObj.lengths[f]
tf_t_d_f = entityObj.term_freqs[f].get(qt,0)
tf_t_C_f = lucene_obj.get_coll_termfreq(qt, f)
tempSim=get_dirichlet_prob(tf_t_d_f, len_d_f, tf_t_C_f, len_C_f[f], mu[f])
# compute f(p(t1|De),p(t2|De)...)
localSim+=mlm_weights[f]*tempSim
if localSim>0.0:
totalSim+=math.log(localSim)
return totalSim
def sdmSim(queryObj,entityObj,field,lucene_obj):
ft=fo=fu=0.0
len_C_f = lucene_obj.get_coll_length(field)
mu=lucene_obj.get_avg_len(field)
ft=lmSim(queryObj.contents_obj,entityObj,field,None,'log_sum',lucene_obj)
if LAMBDA_O>0:
for bigram_pair in queryObj.bigrams:
bigram=bigram_pair[0]+' '+bigram_pair[1]
tf,cf=lucene_obj.get_coll_bigram_freq(bigram,field,True,0,entityObj.title)
ptd=get_dirichlet_prob(tf,entityObj.length,cf,len_C_f,mu)
if ptd>0:
fo+=math.log(ptd)
if LAMBDA_U>0:
for bigram_pair in queryObj.bigrams:
bigram=bigram_pair[0]+' '+bigram_pair[1]
tf,cf=lucene_obj.get_coll_bigram_freq(bigram,field,False,6,entityObj.title)
ptd=get_dirichlet_prob(tf,entityObj.length,cf,len_C_f,mu)
if ptd>0:
fu+=math.log(ptd)
score=LAMBDA_T*ft+LAMBDA_O*fo+LAMBDA_U*fu
return score
def fsdmSim(queryObj,entityObj,lucene_obj):
fields=LIST_F
len_C_f={}
mu={}
for f in LIST_F:
len_C_f[f]=lucene_obj.get_coll_length(f)
mu[f]=lucene_obj.get_avg_len(f)
ft=fo=fu=0.0
# w is a dict of weights for each field
# compute ft
for t in queryObj.contents_obj.term:
w=get_mapping_prob(t,lucene_obj)
ft_temp=0.0
for field in w:
tf=entityObj.term_freqs[field].get(t,0)
cf=lucene_obj.get_coll_termfreq(t, field)
ptd=get_dirichlet_prob(tf,entityObj.lengths[field],cf,len_C_f[field],mu[field])
if ptd>0:
ft_temp+=ptd*w[field]
if ft_temp>0:
ft+=math.log(ft_temp)
# compute fo
if LAMBDA_O>0:
for bigram_pair in queryObj.bigrams:
bigram=bigram_pair[0]+' '+bigram_pair[1]
w=get_mapping_prob(bigram,lucene_obj,ordered=True,slop=0)
fo_temp=0.0
for field in w:
tf,cf=lucene_obj.get_coll_bigram_freq(bigram,field,True,0,entityObj.title)
ptd=get_dirichlet_prob(tf,entityObj.lengths[field],cf,len_C_f[field],mu[field])
if ptd>0:
fo_temp+=ptd*w[field]
if fo_temp>0:
fo+=math.log(fo_temp)
# compute fu
if LAMBDA_U>0:
for bigram_pair in queryObj.bigrams:
bigram=bigram_pair[0]+' '+bigram_pair[1]
w=get_mapping_prob(bigram,lucene_obj,ordered=False,slop=6)
fu_temp=0.0
for field in w:
tf,cf=lucene_obj.get_coll_bigram_freq(bigram,field,False,6,entityObj.title)
ptd=get_dirichlet_prob(tf,entityObj.lengths[field],cf,len_C_f[field],mu[field])
if ptd>0:
fu_temp+=ptd*w[field]
if fu_temp>0:
fu+=math.log(fu_temp)
'''
if queryObj.contents_obj.length>1:
ft/=queryObj.contents_obj.length
fo/=(queryObj.contents_obj.length-1)
fu/=(queryObj.contents_obj.length-1)
'''
score=LAMBDA_T*ft+LAMBDA_O*fo+LAMBDA_U*fu
return score
def get_mapping_prob(t,lucene_obj,ordered=True,slop=0):
"""
Computes PRMS field mapping probability.
p(f|t) = P(t|f)P(f) / sum_f'(P(t|C_{f'_c})P(f'))
:param t: str
:param coll_termfreq_fields: {field: freq, ...}
:return Dictionary {field: prms_prob, ...}
"""
fields=LIST_F
if len(fields)==1:
# for sdm and lm
return {fields[0]:1.0}
is_bigram=True if t.find(' ')>-1 else False
#find cache
coll_termfreq_fields={}
for f in fields:
if is_bigram==False:
coll_termfreq_fields[f]=lucene_obj.get_coll_termfreq(t, f)
else:
coll_termfreq_fields[f]=lucene_obj.get_coll_bigram_freq(t,f,ordered,slop,'NONE')[1]
total_field_freq=lucene_obj.get_total_field_freq(fields)
# calculates numerators for all fields: P(t|f)P(f)
numerators = {}
for f in fields:
p_t_f = coll_termfreq_fields[f] / lucene_obj.get_coll_length(f)
p_f = lucene_obj.get_doc_count(f) / total_field_freq
p_f_t = p_t_f * p_f
if p_f_t > 0:
numerators[f] = p_f_t
else:
numerators[f]=0
# calculates denominator: sum_f'(P(t|C_{f'_c})P(f'))
denominator = sum(numerators.values())
mapping_probs = {}
if denominator > 0: # if the term is present in the collection
for f in numerators:
mapping_probs[f] = numerators[f] / denominator
return mapping_probs
# ============================
def mlm_sas(queryObj,entityObj,structure,lucene_handler):
if len(entityObj.categories)==0:
return NEGATIVE_INFINITY
D=structure.cat_dag
lucene_cat=lucene_handler['category_corpus']
lucene_doc=lucene_handler['first_pass']
len_d_f=entityObj.lengths
sum_score=0.0
max_score=NEGATIVE_INFINITY
len_C_f={}
mlm_weights={}
sum_ptc={}
mu={}
for field in LIST_F:
len_C_f[field]=lucene_doc.get_coll_length(field)
mu[field]=lucene_doc.get_avg_len(field)
mlm_weights[field]=1.0/len(LIST_F)
sum_ptc[field]=[0.0 for i in range(queryObj.contents_obj.length)]
if MODEL_NAME=='mlm-tc':
mlm_weights={'stemmed_names':0.2,'stemmed_catchall':0.8} if USED_QUERY_VERSION=='stemmed_raw_query' else {'names':0.2,'attributes':0.8}
curPath=[]
def smooth_path(cat,path_len,alpha_t,sum_nominator):
nonlocal D,curPath,sum_ptc,cnt_path
nonlocal max_score_p_cat,max_score
nonlocal lucene_cat,lucene_doc
if cnt_path>TOP_PATH_NUM_PER_CAT:
return
# the following is end condition
if path_len==LIMIT_SAS_PATH_LENGTH or len(D[cat])==0:
# compute score
cnt_path+=1
if alpha_t==ALPHA_SAS:
return
cof=(1-ALPHA_SAS)/(ALPHA_SAS-alpha_t)
#cof=0.003
score_p=0.0
for j in range(queryObj.contents_obj.length):
term=queryObj.contents_obj.term[j]
ptd=0.0
for f in LIST_F:
tf_d_f=entityObj.term_freqs[f].get(term,0.0)
cf_f = lucene_doc.get_coll_termfreq(term, f)
ptc_doc=cf_f/len_C_f[f] if len_C_f[f]>0 else 0.0
ptd_f=(tf_d_f+mu[f]*ptc_doc+cof*sum_ptc[f][j])/(len_d_f[f]+mu[f]+cof*sum_nominator[f]) if (len_d_f[f]+mu[f]+cof*sum_nominator[f])>0 else 0.0
'''
if tf_d_f>0:
ptd_f=(tf_d_f+mu[f]*ptc_doc+cof*sum_ptc[f][j])/(len_d_f[f]+mu[f]+cof*sum_nominator[f]) if len_d_f[f]+mu[f]+cof*sum_nominator[f]>0 else 0.0
else:
ptd_f=(tf_d_f+mu[f]*ptc_doc)/(len_d_f[f]+mu[f]) if len_d_f[f]+mu[f]>0 else 0.0
'''
ptd+=mlm_weights[f]*ptd_f
if ptd>0:
score_p+=math.log(ptd)
if score_p>max_score_p_cat:
max_score_p_cat=score_p
return
# maintain useful temporary variables
d,docID=lucene_cat.findDoc(cat,'category',True)
bak_sum_ptc=sum_ptc.copy()
if d is not None:
# maintain
cnt_doc_corpus=int(d['num_articles'])
for f in LIST_F:
# get category corpus
term_freq=lucene_cat.get_term_freq(docID,f,True)
len_c=sum(term_freq.values())
mu_c=len_c/cnt_doc_corpus if cnt_doc_corpus>0 else 0.0
sum_nominator[f]+=alpha_t*mu_c
for j in range(queryObj.contents_obj.length):
term=queryObj.contents_obj.term[j]
tf_c=term_freq.get(term,0.0)
ptc=tf_c/len_c if len_c>0 else -1
if ptc>-1:
sum_ptc[f][j]+=(alpha_t*mu_c*ptc)
cnt=0
for child in iter(D[cat]):
if child in D:
curPath.append(child)
smooth_path(child,path_len+1,alpha_t*ALPHA_SAS,sum_nominator)
curPath.pop()
sum_ptc=bak_sum_ptc.copy()
cnt+=1
if cnt>TOP_CATEGORY_NUM:
break
# end of function smooth_path
for cat in entityObj.categories[:TOP_CATEGORY_NUM]:
if cat not in D:
continue
max_score_p_cat=NEGATIVE_INFINITY
cnt_path=0
smooth_path(cat,1,ALPHA_SAS,{f:0.0 for f in LIST_F})
if max_score_p_cat>NEGATIVE_INFINITY:
sum_score+=max_score_p_cat
if max_score_p_cat>max_score:
max_score=max_score_p_cat
return max_score
#===========================================
def lm_sas(queryObj,entityObj,structure,lucene_handler,mongoObj,field):
if len(entityObj.categories)==0:
return NEGATIVE_INFINITY
D=structure.cat_dag
lucene_cat=lucene_handler['category_corpus']
lucene_doc=lucene_handler['first_pass']
termList=entityObj.term_freq
len_d=entityObj.length
sum_score=0.0
max_score=NEGATIVE_INFINITY
len_C_f = lucene_doc.get_coll_length(field)
mu_d=lucene_doc.get_avg_len(field)
curPath=[]
sum_ptc=[0.0 for i in range(queryObj.contents_obj.length)]
def smooth_path(cat,path_len,alpha_t,sum_nominator):
nonlocal D,curPath,sum_ptc,cnt_path
nonlocal max_score_p_cat,max_score
nonlocal lucene_cat,lucene_doc
if cnt_path>TOP_PATH_NUM_PER_CAT:
return
if path_len==LIMIT_SAS_PATH_LENGTH or len(D[cat])==0:
# compute score
cnt_path+=1
if alpha_t==ALPHA_SAS:
return
cof=(1-ALPHA_SAS)/(ALPHA_SAS-alpha_t)
#cof=0.003
score_p=0.0
for j in range(queryObj.contents_obj.length):
term=queryObj.contents_obj.term[j]
tf_d=entityObj.term_freq.get(term,0.0)
tf_t_C_f = lucene_doc.get_coll_termfreq(term, field)
ptc_doc=tf_t_C_f/len_C_f if len_C_f>0 else 0.0
ptd=(tf_d+mu_d*ptc_doc+cof*sum_ptc[j])/(len_d+mu_d+cof*sum_nominator) if len_d+mu_d+cof*sum_nominator>0 else 0.0
'''
if tf_d==0:
ptd=(tf_d+mu_d*ptc_doc+cof*sum_ptc[j])/(len_d+mu_d+cof*sum_nominator) if len_d+mu_d+cof*sum_nominator>0 else 0.0
else:
ptd=(tf_d+mu_d*ptc_doc)/(len_d+mu_d) if len_d+mu_d>0 else 0.0
'''
if ptd>0:
score_p+=math.log(ptd)
if score_p>max_score_p_cat:
max_score_p_cat=score_p
return
# maintain useful temporary variables
d,docID=lucene_cat.findDoc(cat,'category',True)
bak_sum_ptc=sum_ptc[:]
if d is not None:
# maintain
term_freq=lucene_cat.get_term_freq(docID,field,True)
len_c=sum(term_freq.values())
cnt_doc_corpus=int(d['num_articles'])
mu_c=len_c/cnt_doc_corpus if cnt_doc_corpus>0 else 0.0
sum_nominator+=alpha_t*mu_c
for j in range(queryObj.contents_obj.length):
term=queryObj.contents_obj.term[j]
tf_c=term_freq.get(term,0.0)
ptc=tf_c/len_c if len_c>0 else -1
if ptc>-1:
sum_ptc[j]+=(alpha_t*ptc*mu_c)
cnt=0
for child in iter(D[cat]):
cnt+=1
if cnt>TOP_CATEGORY_NUM:
break
if child in D:
curPath.append(child)
smooth_path(child,path_len+1,alpha_t*ALPHA_SAS,sum_nominator)
curPath.pop()
sum_ptc=bak_sum_ptc[:]
# end of function smooth_path
for cat in entityObj.categories[:TOP_CATEGORY_NUM]:
if cat not in D:
continue
max_score_p_cat=NEGATIVE_INFINITY
cnt_path=0
smooth_path(cat,1,ALPHA_SAS,0.0)
if max_score_p_cat>NEGATIVE_INFINITY:
sum_score+=max_score_p_cat
if max_score_p_cat>max_score:
max_score=max_score_p_cat
return max_score
# ============================