-
Notifications
You must be signed in to change notification settings - Fork 21
/
algo.py
1646 lines (1400 loc) · 65.2 KB
/
algo.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
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
import matplotlib.pyplot as plt
from sklearn import metrics
import random
import math
import datetime
import uuid
from Levenshtein import ratio
from fingerprint import Fingerprint
from sklearn.ensemble import RandomForestClassifier
from sklearn.externals import joblib
import numpy as np
import string
from multiprocessing import Pool, Pipe
import time
results = []
def generate_replay_sequence(fp_set, visit_frequency):
"""
Takes as input a set of fingerprint fp_set,
a frequency of visit visit_frequency in days
Returns a list of fingerprints in the order they must be replayed
"""
# we start by generating the sequence of replay
# we don't store the last fp of each user since it's not realistic to replay it infinitely
user_id_to_fps = dict()
for fingerprint in fp_set:
if fingerprint.getId() not in user_id_to_fps:
user_id_to_fps[fingerprint.getId()] = []
user_id_to_fps[fingerprint.getId()].append(fingerprint)
user_id_to_sequence = dict()
for user_id in user_id_to_fps:
# can be removed later when we don't set a limit on counter
if len(user_id_to_fps[user_id]) > 1:
user_id_to_fps[user_id] = user_id_to_fps[user_id][:-1]
sequence = []
last_visit = user_id_to_fps[user_id][0].getStartTime()
counter_suffix = "i"
assigned_counter = "%d_%s" % (user_id_to_fps[user_id][0].getCounter(), counter_suffix)
sequence.append((assigned_counter, last_visit))
for fingerprint in user_id_to_fps[user_id]:
counter_suffix = 0
# if it is none and not the last one (last one is removed)
# it means the fp changed within the same time interval
if fingerprint.getEndTime() is not None:
while last_visit + datetime.timedelta(days=visit_frequency) < \
fingerprint.getEndTime():
last_visit = last_visit + datetime.timedelta(days=visit_frequency)
assigned_counter = "%d_%d" % (fingerprint.getCounter(), counter_suffix)
sequence.append((assigned_counter, last_visit))
counter_suffix += 1
user_id_to_sequence[user_id] = sequence
# now we generate the whole sequence
# we start by merging all the subsequences, and then sort it by the date
replay_sequence = []
for user_id in user_id_to_sequence:
replay_sequence += user_id_to_sequence[user_id]
replay_sequence = sorted(replay_sequence, key=lambda x: x[1])
return replay_sequence
def split_data(perc_train, fingerprint_dataset):
"""
Takes as input the percentage of fingerprints for training and
the fingerprint dataset ordered chronologically.
Returns the training and the test sequence
"""
index_split = int(len(fingerprint_dataset) * perc_train)
# train, test
return fingerprint_dataset[: index_split], fingerprint_dataset[index_split:]
def generate_new_id():
"""
Returns a random user id
"""
return str(uuid.uuid4())
def candidates_have_same_id(candidate_list):
"""
Returns True if all candidates have the same id
Else False
"""
if len(candidate_list) == 0:
return False
return not any(not x for x in [y[2] == candidate_list[0][2] for y in candidate_list])
def rule_based(fingerprint_unknown, user_id_to_fps, counter_to_fingerprint):
"""
Given an unknown fingerprint fingerprint_unknown,
and a set of known fingerprints fps_available,
tries to link fingerprint_unknown to a fingerprint in
fps_available.
If it can be linked it returns the id of the fingerprint it has been linked with,
otherwise it returns a new generated user id.
"""
forbidden_changes = [
Fingerprint.CANVAS_JS_HASHED,
Fingerprint.LOCAL_JS,
Fingerprint.DNT_JS,
Fingerprint.COOKIES_JS
]
allowed_changes_with_sim = [
Fingerprint.USER_AGENT_HTTP,
Fingerprint.VENDOR,
Fingerprint.RENDERER,
Fingerprint.PLUGINS_JS,
Fingerprint.LANGUAGE_HTTP,
Fingerprint.ACCEPT_HTTP
]
allowed_changes = [
Fingerprint.RESOLUTION_JS,
Fingerprint.ENCODING_HTTP,
Fingerprint.TIMEZONE_JS
]
ip_allowed = False
candidates = list()
exact_matching = list()
prediction = None
for user_id in user_id_to_fps:
for counter_str in user_id_to_fps[user_id]:
counter_known = int(counter_str.split("_")[0])
fingerprint_known = counter_to_fingerprint[counter_known]
# check fingerprint full hash for exact matching
if fingerprint_known.hash == fingerprint_unknown.hash:
# either we look if there are multiple users that match
# in that case we create new id
# or we assign randomly?
exact_matching.append((counter_str, None, user_id))
elif len(exact_matching) < 1 and fingerprint_known.constant_hash == \
fingerprint_unknown.constant_hash:
# we make the comparison only if same os/browser/platform
if fingerprint_known.val_attributes[Fingerprint.GLOBAL_BROWSER_VERSION] > \
fingerprint_unknown.val_attributes[Fingerprint.GLOBAL_BROWSER_VERSION]:
continue
if fingerprint_known.hasFlashActivated() and fingerprint_unknown.hasFlashActivated() and \
not fingerprint_known.areFontsSubset(fingerprint_unknown):
continue
forbidden_change_found = False
for attribute in forbidden_changes:
if fingerprint_known.val_attributes[attribute] != \
fingerprint_unknown.val_attributes[attribute]:
forbidden_change_found = True
break
if forbidden_change_found:
continue
nb_changes = 0
changes = []
# we allow at most 2 changes, then we check for similarity
for attribute in allowed_changes_with_sim:
if fingerprint_known.val_attributes[attribute] != \
fingerprint_unknown.val_attributes[attribute]:
changes.append(attribute)
nb_changes += 1
if nb_changes > 2:
break
if nb_changes > 2:
continue
sim_too_low = False
for attribute in changes:
if ratio(fingerprint_known.val_attributes[attribute],
fingerprint_unknown.val_attributes[attribute]) < 0.75:
sim_too_low = True
break
if sim_too_low:
continue
nb_allowed_changes = 0
for attribute in allowed_changes:
if fingerprint_known.val_attributes[attribute] != \
fingerprint_unknown.val_attributes[attribute]:
nb_allowed_changes += 1
if nb_allowed_changes > 1:
break
if nb_allowed_changes > 1:
continue
total_nb_changes = nb_allowed_changes + nb_changes
if total_nb_changes == 0:
exact_matching.append((counter_str, None, user_id))
else:
candidates.append((counter_str, total_nb_changes, user_id))
if len(exact_matching) > 0:
if len(exact_matching) == 1 or candidates_have_same_id(exact_matching):
return exact_matching[0][2]
elif ip_allowed:
# we don't use IP address, it is just here because of a previous test!
for elt in exact_matching:
counter = int(elt[0].split("_")[0])
fingerprint_known = counter_to_fingerprint[counter_known]
if fingerprint_known.val_attributes[Fingerprint.ADDRESS_HTTP] == \
fingerprint_unknown.val_attributes[Fingerprint.ADDRESS_HTTP]:
prediction = elt[2]
break
else:
if len(candidates) == 1 or candidates_have_same_id(candidates):
prediction = candidates[0][2]
elif ip_allowed:
# we don't use IP address, it is just here because of a previous test!
for elt in candidates:
counter = int(elt[0].split("_")[0])
fingerprint_known = counter_to_fingerprint[counter_known]
if fingerprint_known.val_attributes[Fingerprint.ADDRESS_HTTP] == \
fingerprint_unknown.val_attributes[Fingerprint.ADDRESS_HTTP]:
prediction = elt[2]
break
if prediction is None:
prediction = generate_new_id()
return prediction
def simple_eckersley(fingerprint_unknown, user_id_to_fps, counter_to_fingerprint):
"""
Given an unknown fingerprint fingerprint_unknown,
and a set of known fingerprints fps_available,
tries to link fingerprint_unknown to a fingerprint in
fps_available.
If it can be linked it returns the id of the fingerprint it has been linked with,
otherwise it returns a new generated user id.
"""
# order of attributes matter, should place most discriminative first to decrease average
# number of comparisons
attributes_to_test = ["fontsFlashHashed", "pluginsJSHashed", "userAgentHttp", "resolutionJS", "acceptHttp",
"timezoneJS", "cookiesJS", "localJS"]
candidates = list()
exact = False
for user_id in user_id_to_fps:
for counter_str in user_id_to_fps[user_id]:
attributes_different = 0
modified_attribute = ""
counter_known = int(counter_str.split("_")[0])
fingerprint_known = counter_to_fingerprint[counter_known]
for attribute in attributes_to_test:
# special case for Flash fonts
if attribute == Fingerprint.FONTS_FLASH_HASHED:
# we consider that flash activation/deactivation is not a difference
if fingerprint_known.hasFlashActivated() and \
fingerprint_unknown.hasFlashActivated():
if fingerprint_known.val_attributes[attribute] != \
fingerprint_unknown.val_attributes[attribute]:
attributes_different += 1
modified_attribute = attribute
elif fingerprint_unknown.val_attributes[attribute] != \
fingerprint_known.val_attributes[attribute]:
attributes_different += 1
modified_attribute = attribute
if attributes_different > 1:
break
if attributes_different == 1:
# (new_counter, modified_attribute, assigned_id)
candidates.append((counter_str, modified_attribute, user_id))
elif attributes_different == 0:
prediction = user_id
exact = True
if len(candidates) == 1 or candidates_have_same_id(candidates):
if candidates[0][1] in ["cookiesJS", "resolutionJS", "timezoneJS", "IEDataJS",
"localJS", "dntJS"]:
prediction = candidates[0][2]
else:
counter_to_test = int(candidates[0][0].split("_")[0])
ratio_sim = ratio(counter_to_fingerprint[counter_to_test].val_attributes[candidates[0][1]],
fingerprint_unknown.val_attributes[candidates[0][1]])
if ratio_sim > 0.85:
prediction = candidates[0][2]
else:
prediction = generate_new_id()
elif not exact:
prediction = generate_new_id()
return prediction
def replay_scenario(fingerprint_dataset, visit_frequency, link_fingerprint, \
filename="./results/scenario_replay_result.csv", model=None, lambda_threshold=None):
"""
Takes as input the fingerprint dataset,
the frequency of visit in days,
link_fingerprint, the function used for the linking strategy
filename, path to the file to save results of the scenario
"""
nb_max_cmp = 2
replay_sequence = generate_replay_sequence(fingerprint_dataset, visit_frequency)
counter_to_fingerprint = dict()
for fingerprint in fingerprint_dataset:
counter_to_fingerprint[fingerprint.getCounter()] = fingerprint
fps_available = [] # set of known fingerprints (new_counter, new_id)
user_id_to_fps = dict()
counter_to_time = dict()
for index, elt in enumerate(replay_sequence):
if index % 500 == 0:
print(index)
counter_to_time[elt[0]] = elt[1]
counter = int(elt[0].split("_")[0])
fingerprint_unknown = counter_to_fingerprint[counter]
if model is None:
assigned_id = link_fingerprint(fingerprint_unknown, user_id_to_fps, \
counter_to_fingerprint)
else:
assigned_id = link_fingerprint(fingerprint_unknown, user_id_to_fps, \
counter_to_fingerprint, model, lambda_threshold)
fps_available.append((elt[0], assigned_id))
if assigned_id not in user_id_to_fps:
user_id_to_fps[assigned_id] = []
elif len(user_id_to_fps[assigned_id]) == nb_max_cmp:
user_id_to_fps[assigned_id].pop(0)
user_id_to_fps[assigned_id].append(elt[0])
# every 2000 elements we delete elements too old
if index % 2000 == 0:
# 40 days in seconds
time_limit = 30 * 24 * 60 * 60
ids_to_remove = set()
current_time = elt[1]
for user_id in user_id_to_fps:
counter_str = user_id_to_fps[user_id][-1]
time_tmp = counter_to_time[counter_str]
if (current_time - time_tmp).total_seconds() > time_limit:
ids_to_remove.add(user_id)
for user_id in ids_to_remove:
del user_id_to_fps[user_id]
with open(filename, "w") as f:
for elt in fps_available:
f.write("%s,%s\n" % (elt[0], elt[1]))
return fps_available
def generateHeader(attributes):
header = []
for attribute in attributes:
if attribute == Fingerprint.ID:
pass
elif attribute == Fingerprint.CREATION_TIME:
header.append(attribute)
elif attribute == Fingerprint.ENCODING_HTTP:
header.append(attribute)
elif attribute == Fingerprint.TIMEZONE_JS:
header.append(attribute)
elif attribute == Fingerprint.PLUGINS_JS:
header.append("simPlugs")
elif attribute == Fingerprint.RESOLUTION_JS:
header.append(attribute)
elif attribute == Fingerprint.CANVAS_JS_HASHED:
header.append(attribute)
elif attribute == Fingerprint.FONTS_FLASH:
header.append("hasFlash")
header.append("sameFonts")
else:
header.append(attribute)
header.append("nbChange")
return header
def compute_similarity_fingerprint(fp1, fp2, attributes, train_mode):
similarity_vector = []
flash_activated = fp1.hasFlashActivated() and fp2.hasFlashActivated()
nb_changes = 0
for attribute in attributes:
if attribute == Fingerprint.ID:
val_to_insert = (1 if fp1.belongToSameUser(fp2) else 0)
similarity_vector.insert(0, val_to_insert)
elif attribute == Fingerprint.CREATION_TIME:
diff = fp1.getTimeDifference(fp2)
similarity_vector.append(diff)
elif attribute == Fingerprint.ENCODING_HTTP:
similarity_vector.append(1) if fp1.hasSameEncodingHttp(fp2) else similarity_vector.append(0)
elif attribute == Fingerprint.TIMEZONE_JS:
similarity_vector.append(1) if fp1.hasSameTimezone(fp2) else similarity_vector.append(0)
elif attribute == Fingerprint.PLUGINS_JS:
sim = ratio(fp1.val_attributes[attribute], fp2.val_attributes[attribute])
similarity_vector.append(sim)
elif attribute == Fingerprint.RESOLUTION_JS:
similarity_vector.append(1) if fp1.hasSameResolution(fp2) else similarity_vector.append(0)
elif attribute == Fingerprint.CANVAS_JS_HASHED:
similarity_vector.append(1) if fp1.hasSameCanvasJsHashed(fp2) else similarity_vector.append(0)
elif attribute == Fingerprint.FONTS_FLASH:
if flash_activated:
similarity_vector.append(1)
similarity_vector.append(1) if fp1.hasSameFonts(fp2) else similarity_vector.append(0)
else:
similarity_vector.append(0)
similarity_vector.append(0)
else:
sim = ratio(str(fp1.val_attributes[attribute]), str(fp2.val_attributes[attribute]))
similarity_vector.append(sim)
if fp1.val_attributes[attribute] != fp2.val_attributes[attribute]:
nb_changes += 1
if nb_changes > 5 and not train_mode:
return None, None
similarity_vector.append(nb_changes)
return np.asarray(similarity_vector[1:]), np.asarray(similarity_vector[0])
def train_ml(fingerprint_dataset, train_data, load=True, \
model_path="./data/my_ml_model"):
if load:
model = joblib.load(model_path)
else:
counter_to_fingerprint = dict()
index_to_user_id = dict()
user_ids = set()
index = 0
not_to_test = set([Fingerprint.PLATFORM_FLASH,
Fingerprint.PLATFORM_INCONSISTENCY,
Fingerprint.PLATFORM_JS,
Fingerprint.PLUGINS_JS_HASHED,
Fingerprint.SESSION_JS,
Fingerprint.IE_DATA_JS,
Fingerprint.ADDRESS_HTTP,
Fingerprint.BROWSER_FAMILY,
Fingerprint.COOKIES_JS,
Fingerprint.DNT_JS,
Fingerprint.END_TIME,
Fingerprint.FONTS_FLASH_HASHED,
Fingerprint.GLOBAL_BROWSER_VERSION,
Fingerprint.LANGUAGE_FLASH,
Fingerprint.LANGUAGE_INCONSISTENCY,
Fingerprint.LOCAL_JS,
Fingerprint.MINOR_BROWSER_VERSION,
Fingerprint.MAJOR_BROWSER_VERSION,
Fingerprint.NB_FONTS,
Fingerprint.NB_PLUGINS,
Fingerprint.COUNTER,
Fingerprint.OS,
Fingerprint.ACCEPT_HTTP,
Fingerprint.CONNECTION_HTTP,
Fingerprint.ENCODING_HTTP,
Fingerprint.RESOLUTION_FLASH,
Fingerprint.TIMEZONE_JS,
Fingerprint.VENDOR,
])
att_ml = set(fingerprint_dataset[0].val_attributes.keys())
att_ml = sorted([x for x in att_ml if x not in not_to_test])
for fingerprint in fingerprint_dataset:
counter_to_fingerprint[fingerprint.getCounter()] = fingerprint
if fingerprint.getId() not in user_ids:
user_ids.add(fingerprint.getId())
index_to_user_id[index] = fingerprint.getId()
index += 1
# just to simplify negative comparisons later
# we generate multiple replay sequences on train data with different visit frequencies
# to generate more diverse training data
print("Start generating training data")
for visit_frequency in range(1, 10):
print(visit_frequency)
train_replay_sequence = generate_replay_sequence(train_data, visit_frequency)
# we group fingerprints by user id
user_id_to_fps = dict()
for elt in train_replay_sequence:
counter = int(elt[0].split("_")[0])
fingerprint = counter_to_fingerprint[counter]
if fingerprint.getId() not in user_id_to_fps:
user_id_to_fps[fingerprint.getId()] = []
user_id_to_fps[fingerprint.getId()].append(fingerprint)
# we generate the training data
X, y = [], []
attributes = sorted(fingerprint_dataset[0].val_attributes.keys())
for user_id in user_id_to_fps:
previous_fingerprint = None
for fingerprint in user_id_to_fps[user_id]:
if previous_fingerprint is not None:
x_row, y_row = compute_similarity_fingerprint(fingerprint, previous_fingerprint, att_ml,
train_mode=True)
X.append(x_row)
y.append(y_row)
previous_fingerprint = fingerprint
# we compute negative rows
for user_id in user_id_to_fps:
for fp1 in user_id_to_fps[user_id]:
try:
compare_with_id = index_to_user_id[random.randint(0, len(user_id_to_fps))]
compare_with_fp = random.randint(0, len(user_id_to_fps[compare_with_id]))
fp2 = user_id_to_fps[compare_with_id][compare_with_fp]
x_row, y_row = compute_similarity_fingerprint(fp1, fp2, att_ml, train_mode=True)
X.append(x_row)
y.append(y_row)
except:
pass
print("Start training model")
model = RandomForestClassifier(n_jobs=4)
print("Training data: %d" % len(X))
model.fit(X, y)
print("Model trained")
joblib.dump(model, model_path)
print("model saved at: %s" % model_path)
return model
def ml_based(fingerprint_unknown, user_id_to_fps, counter_to_fingerprint, model, lambda_threshold):
forbidden_changes = [
Fingerprint.LOCAL_JS,
Fingerprint.DNT_JS,
Fingerprint.COOKIES_JS
]
allowed_changes_with_sim = [
Fingerprint.USER_AGENT_HTTP,
Fingerprint.VENDOR,
Fingerprint.RENDERER,
Fingerprint.PLUGINS_JS,
Fingerprint.LANGUAGE_HTTP,
Fingerprint.ACCEPT_HTTP
]
allowed_changes = [
Fingerprint.RESOLUTION_JS,
Fingerprint.ENCODING_HTTP,
]
not_to_test = set([Fingerprint.PLATFORM_FLASH,
Fingerprint.PLATFORM_INCONSISTENCY,
Fingerprint.PLATFORM_JS,
Fingerprint.PLUGINS_JS_HASHED,
Fingerprint.SESSION_JS,
Fingerprint.IE_DATA_JS,
Fingerprint.ADDRESS_HTTP,
Fingerprint.BROWSER_FAMILY,
Fingerprint.COOKIES_JS,
Fingerprint.DNT_JS,
Fingerprint.END_TIME,
Fingerprint.FONTS_FLASH_HASHED,
Fingerprint.GLOBAL_BROWSER_VERSION,
Fingerprint.LANGUAGE_FLASH,
Fingerprint.LANGUAGE_INCONSISTENCY,
Fingerprint.LOCAL_JS,
Fingerprint.MINOR_BROWSER_VERSION,
Fingerprint.MAJOR_BROWSER_VERSION,
Fingerprint.NB_FONTS,
Fingerprint.NB_PLUGINS,
Fingerprint.COUNTER,
Fingerprint.OS,
Fingerprint.ACCEPT_HTTP,
Fingerprint.CONNECTION_HTTP,
Fingerprint.ENCODING_HTTP,
Fingerprint.RESOLUTION_FLASH,
Fingerprint.TIMEZONE_JS,
Fingerprint.VENDOR,
])
att_ml = set(fingerprint_unknown.val_attributes.keys())
att_ml = sorted([x for x in att_ml if x not in not_to_test])
ip_allowed = False
candidates = list()
exact_matching = list()
prediction = None
for user_id in user_id_to_fps:
for counter_str in user_id_to_fps[user_id]:
counter_known = int(counter_str.split("_")[0])
fingerprint_known = counter_to_fingerprint[counter_known]
# check fingerprint full hash for exact matching
if fingerprint_known.hash == fingerprint_unknown.hash:
exact_matching.append((counter_str, None, user_id))
elif len(exact_matching) < 1 and fingerprint_known.constant_hash == \
fingerprint_unknown.constant_hash:
# we make the comparison only if same os/browser/platform
if fingerprint_known.val_attributes[Fingerprint.GLOBAL_BROWSER_VERSION] > \
fingerprint_unknown.val_attributes[Fingerprint.GLOBAL_BROWSER_VERSION]:
continue
forbidden_change_found = False
for attribute in forbidden_changes:
if fingerprint_known.val_attributes[attribute] != \
fingerprint_unknown.val_attributes[attribute]:
forbidden_change_found = True
break
if forbidden_change_found:
continue
candidates.append((counter_str, None, user_id))
if len(exact_matching) > 0:
if len(exact_matching) == 1 or candidates_have_same_id(exact_matching):
return exact_matching[0][2]
elif len(candidates) > 0:
# in this case we apply ML
data = []
attributes = sorted(fingerprint_unknown.val_attributes.keys())
new_candidates = []
for elt in candidates:
counter = int(elt[0].split("_")[0])
fingerprint_known = counter_to_fingerprint[counter]
x_row, _ = compute_similarity_fingerprint(fingerprint_unknown,
fingerprint_known,
att_ml, train_mode=False)
if x_row is not None:
data.append(x_row)
new_candidates.append(elt)
if len(new_candidates) > 0:
predictions_model = model.predict_proba(data)
predictions_model = 1.0 - predictions_model
nearest = (-predictions_model[:, 0]).argsort()[:3]
max_nearest = 1
second_proba = None
for i in range(1, len(nearest)):
if predictions_model[nearest[i], 0] != predictions_model[nearest[0], 0]:
max_nearest = i
second_proba = predictions_model[nearest[i], 0]
break
nearest = nearest[:max_nearest]
diff_enough = True
if second_proba is not None and predictions_model[nearest[0], 0] < second_proba + 0.1: # 0.1 = diff parameter
diff_enough = False
if diff_enough and predictions_model[nearest[0], 0] > lambda_threshold and candidates_have_same_id(
[candidates[x] for x in nearest]):
prediction = new_candidates[nearest[0]][2]
if prediction is None:
prediction = generate_new_id()
return prediction
def load_scenario_result(filename):
"""
Loads and returns a scenario result from disk
"""
scenario_result = []
with open(filename, "r") as f:
for line in f:
l_split = line.split(",")
scenario_result.append((l_split[0], l_split[1]))
return scenario_result
def compute_ownership(fingerprints):
real_user_id_to_count = dict()
for fingerprint in fingerprints:
if fingerprint.getId() in real_user_id_to_count:
real_user_id_to_count[fingerprint.getId()] += 1
else:
real_user_id_to_count[fingerprint.getId()] = 1
max_key = max(real_user_id_to_count, key=real_user_id_to_count.get)
return float(real_user_id_to_count[max_key] / len(fingerprints)), max_key
def find_longest_chain(real_user_id, real_id_to_assigned_ids, assigned_ids_to_fingerprint):
"""
For a given user id, tries to find its longest chain
"""
assigned_ids = real_id_to_assigned_ids[real_user_id]
assigned_id_to_count = dict()
for assigned_id in assigned_ids:
tmp_count = 0
for fingerprint in assigned_ids_to_fingerprint[assigned_id]:
if fingerprint.getId() == real_user_id:
tmp_count += 1
assigned_id_to_count[assigned_id] = tmp_count
return max(assigned_id_to_count.items(), key=lambda x: x[1])[1]
def analyse_scenario_result(scenario_result, fingerprint_dataset,
fileres1="./results/res1.csv",
fileres2="./results/res2.csv"):
"""
Performs an analysis of a scenario result
"""
counter_to_fingerprint = dict()
real_user_id_tp_nb_fps = dict()
real_ids = set()
aareal_user_id_to_fps = dict()
for fingerprint in fingerprint_dataset:
counter_to_fingerprint[fingerprint.getCounter()] = fingerprint
real_ids.add(fingerprint.getId())
if fingerprint.getId() not in aareal_user_id_to_fps:
aareal_user_id_to_fps[fingerprint.getId()] = 1
else:
aareal_user_id_to_fps[fingerprint.getId()] += 1
# we map new assigned ids to real ids in database
assigned_ids = set()
real_id_to_assigned_ids = dict()
assigned_id_to_real_ids = dict()
assigned_id_to_fingerprints = dict()
for elt in scenario_result:
counter = int(elt[0].split("_")[0])
assigned_id = elt[1]
assigned_ids.add(assigned_id)
real_db_id = counter_to_fingerprint[counter].getId()
if real_db_id not in real_user_id_tp_nb_fps:
real_user_id_tp_nb_fps[real_db_id] = 1
else:
real_user_id_tp_nb_fps[real_db_id] += 1
if real_db_id not in real_id_to_assigned_ids:
real_id_to_assigned_ids[real_db_id] = set()
real_id_to_assigned_ids[real_db_id].add(assigned_id)
if assigned_id not in assigned_id_to_real_ids:
assigned_id_to_real_ids[assigned_id] = set()
assigned_id_to_fingerprints[assigned_id] = []
assigned_id_to_real_ids[assigned_id].add(counter_to_fingerprint[counter].getId())
assigned_id_to_fingerprints[assigned_id].append(counter_to_fingerprint[counter])
with open(fileres1, "w") as f:
f.write("%s,%s,%s,%s,%s\n" % ("real_id", "nb_assigned_ids", "nb_original_fp", "ratio", "max_chain"))
# don't iterate over reals_ids since some fps don't have end date and are not present
for real_id in real_id_to_assigned_ids:
max_chain = find_longest_chain(real_id, real_id_to_assigned_ids, assigned_id_to_fingerprints)
ratio_stats = real_user_id_tp_nb_fps[real_id] / len(real_id_to_assigned_ids[real_id])
f.write("%s,%d,%d,%f,%d\n" % (real_id,
len(real_id_to_assigned_ids[real_id]),
real_user_id_tp_nb_fps[real_id],
ratio_stats, max_chain)
)
with open(fileres2, "w") as f:
f.write("%s,%s,%s,%s,%s\n" % ("assigned_id", "nb_assigned_ids", "nb_fingerprints",
"ownership", "id_ownership"))
for assigned_id in assigned_id_to_real_ids:
ownership, ownsership_id = compute_ownership(assigned_id_to_fingerprints[assigned_id])
f.write("%s,%d,%d,%f,%s\n" % (assigned_id, len(assigned_id_to_real_ids[assigned_id]),
len(assigned_id_to_fingerprints[assigned_id]), ownership,
ownsership_id))
def compute_distance_top_left(tpr, fp):
return (0 - fp) * (0 - fp) + (1 - tpr) * (1 - tpr)
def optimize_lambda(fingerprint_dataset, train_data, test_data):
counter_to_fingerprint = dict()
index_to_user_id = dict()
user_ids = set()
index = 0
for fingerprint in fingerprint_dataset:
counter_to_fingerprint[fingerprint.getCounter()] = fingerprint
if fingerprint.getId() not in user_ids:
user_ids.add(fingerprint.getId())
index_to_user_id[index] = fingerprint.getId()
index += 1
print("Start generating training data")
for visit_frequency in range(1, 10):
print(visit_frequency)
train_replay_sequence = generate_replay_sequence(train_data, visit_frequency)
# we group fingerprints by user id
user_id_to_fps = dict()
for elt in train_replay_sequence:
counter = int(elt[0].split("_")[0])
fingerprint = counter_to_fingerprint[counter]
if fingerprint.getId() not in user_id_to_fps:
user_id_to_fps[fingerprint.getId()] = []
user_id_to_fps[fingerprint.getId()].append(fingerprint)
# we generate the training data
X, y = [], []
attributes = sorted(fingerprint_dataset[0].val_attributes.keys())
for user_id in user_id_to_fps:
previous_fingerprint = None
for fingerprint in user_id_to_fps[user_id]:
if previous_fingerprint is not None:
x_row, y_row = compute_similarity_fingerprint(fingerprint, previous_fingerprint, attributes,
train_mode=True)
X.append(x_row)
y.append(y_row)
previous_fingerprint = fingerprint
# we compute negative rows
for user_id in user_id_to_fps:
for fp1 in user_id_to_fps[user_id]:
try:
compare_with_id = index_to_user_id[random.randint(0, len(user_id_to_fps))]
compare_with_fp = random.randint(0, len(user_id_to_fps[compare_with_id])-1)
fp2 = user_id_to_fps[compare_with_id][compare_with_fp]
x_row, y_row = compute_similarity_fingerprint(fp1, fp2, attributes, train_mode=True)
X.append(x_row)
y.append(y_row)
except:
print("error")
pass
model = RandomForestClassifier(n_jobs=4)
print("Training data: %d" % len(X))
model.fit(X, y)
print("Finished training")
y_true = []
y_scores = []
for visit_frequency in range(1, 20):
print(visit_frequency)
train_replay_sequence = generate_replay_sequence(test_data, visit_frequency)
# we group fingerprints by user id
user_id_to_fps = dict()
for elt in train_replay_sequence:
counter = int(elt[0].split("_")[0])
fingerprint = counter_to_fingerprint[counter]
if fingerprint.getId() not in user_id_to_fps:
user_id_to_fps[fingerprint.getId()] = []
user_id_to_fps[fingerprint.getId()].append(fingerprint)
attributes = sorted(fingerprint_dataset[0].val_attributes.keys())
x_rows = []
for user_id in user_id_to_fps:
previous_fingerprint = None
for fingerprint in user_id_to_fps[user_id]:
if previous_fingerprint is not None:
x_row, y_row = compute_similarity_fingerprint(fingerprint, previous_fingerprint, attributes, True)
x_rows.append(x_row)
y_true.append(1)
previous_fingerprint = fingerprint
for user_id in user_id_to_fps:
for fp1 in user_id_to_fps[user_id]:
try:
compare_with_id = index_to_user_id[random.randint(0, len(user_id_to_fps))]
compare_with_fp = random.randint(0, len(user_id_to_fps[compare_with_id]))
fp2 = user_id_to_fps[compare_with_id][compare_with_fp]
x_row, y_row = compute_similarity_fingerprint(fp1, fp2, attributes)
x_rows.append(x_row)
y_true.append(0)
except:
pass
predictions = model.predict_proba(x_rows)
for prediction in predictions:
y_scores.append(prediction[1])
fpr, tpr, thresholds = metrics.roc_curve(y_true, y_scores, pos_label=1)
min_indice = 0
min_distance = compute_distance_top_left(tpr[0], fpr[0])
for i in range(1, len(fpr)):
distance = compute_distance_top_left(tpr[i], fpr[i])
if distance < min_distance:
min_indice = i
min_distance = distance
print("best point")
print("%f, %f, %f" % (fpr[min_indice], tpr[min_indice], thresholds[min_indice]))
plt.figure()
lw = 2
plt.plot(fpr, tpr, color='darkorange', lw=lw)
plt.plot([0, 1], [0, 1], color='navy', lw=lw, linestyle='--')
plt.xlim([0.0, 1.0])
plt.ylim([0.0, 1.05])
plt.xlabel('False Positive Rate', fontsize=15)
plt.ylabel('True Positive Rate', fontsize=15)
plt.savefig("./lambda_optim.pdf")
plt.show()
def collect_results(result):
results.extend(result)
def simple_catch(fn, max_diff, nb_cmp_per_id, conn, attributes):
try:
fn(max_diff, nb_cmp_per_id, conn, attributes)
except Exception as e:
print(e)
def candidates_have_same_id_bench(candidate_list):
"""
Returns True if all candidates have the same id
Else False
"""
lf = [x for x in candidate_list if x is not None]
if len(lf) == 0:
return False
return not any(not x for x in [y[0] == lf[0][0] for y in lf])
def parallel_pipe_task_rules_f(max_diff, nb_cmp_per_id, conn, attributes):
forbidden_changes = [
Fingerprint.LOCAL_JS,
Fingerprint.DNT_JS,
Fingerprint.COOKIES_JS
]
allowed_changes_with_sim = [
Fingerprint.USER_AGENT_HTTP,
Fingerprint.VENDOR,
Fingerprint.RENDERER,
Fingerprint.PLUGINS_JS,
Fingerprint.LANGUAGE_HTTP,
Fingerprint.ACCEPT_HTTP
]
allowed_changes = [
Fingerprint.RESOLUTION_JS,
Fingerprint.ENCODING_HTTP,
]
nb_cmp_per_id = 2
user_id_to_fps = dict()
constant_hash_to_user_id = dict()
msg = "CONTINUE"
while msg == "CONTINUE":
fp_to_add = conn.recv()
if fp_to_add == "STOP":
break
if fp_to_add.constant_hash not in constant_hash_to_user_id:
constant_hash_to_user_id[fp_to_add.constant_hash] = set()
constant_hash_to_user_id[fp_to_add.constant_hash].add(fp_to_add.getId())
if fp_to_add.getId() in user_id_to_fps:
user_id_to_fps[fp_to_add.getId()].append(fp_to_add)
else:
user_id_to_fps[fp_to_add.getId()] = list()
user_id_to_fps[fp_to_add.getId()].append(fp_to_add)
msg = conn.recv()
conn.send(len(user_id_to_fps))
print("Finished collecting fps")
msg = "CONTINUE"
print("Start classification process")
avg_nb_cmp = 0
total_nb = 0
while msg == "CONTINUE":
msg = conn.recv()
if msg != "CONTINUE":
break
Xp = []
fingerprint_unknown = conn.recv()
row_index_to_counter = dict()
candidates = list()
exact_matching = list()
prediction = None
if fingerprint_unknown.constant_hash not in constant_hash_to_user_id:
prediction = (generate_new_id(), 1.0)
else:
for user_id in constant_hash_to_user_id[fingerprint_unknown.constant_hash]:
for fingerprint_known in user_id_to_fps[user_id]:
if fingerprint_known.hash == fingerprint_unknown.hash:
# either we look if there are multiple users that match
# in that case we create new id
# or we assign randomly?
exact_matching.append((fingerprint_known, None, user_id))
elif len(exact_matching) < 1 and fingerprint_known.constant_hash == \
fingerprint_unknown.constant_hash:
# we make the comparison only if same os/browser/platform
if fingerprint_known.val_attributes[Fingerprint.GLOBAL_BROWSER_VERSION] > \