File size: 64,422 Bytes
6fa4bc9
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
{
    "paper_id": "2019",
    "header": {
        "generated_with": "S2ORC 1.0.0",
        "date_generated": "2023-01-19T07:29:25.555538Z"
    },
    "title": "Identification of Synthetic Sentence in Bengali News using Hybrid Approach",
    "authors": [
        {
            "first": "Soma",
            "middle": [],
            "last": "Das",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "Indian Institute of Information Technology Kalyani",
                "location": {
                    "addrLine": "West Bengal",
                    "country": "India Sanjay Chatterji"
                }
            },
            "email": ""
        },
        {
            "first": "Rowan",
            "middle": [],
            "last": "Zellers",
            "suffix": "",
            "affiliation": {},
            "email": ""
        },
        {
            "first": "Ari",
            "middle": [],
            "last": "Holtzman",
            "suffix": "",
            "affiliation": {},
            "email": ""
        },
        {
            "first": "Hannah",
            "middle": [],
            "last": "Rashkin",
            "suffix": "",
            "affiliation": {},
            "email": ""
        },
        {
            "first": "Yonatan",
            "middle": [],
            "last": "Bisk",
            "suffix": "",
            "affiliation": {},
            "email": ""
        },
        {
            "first": "Ali",
            "middle": [],
            "last": "Farhadi",
            "suffix": "",
            "affiliation": {},
            "email": ""
        },
        {
            "first": "Franziska",
            "middle": [],
            "last": "Roesner",
            "suffix": "",
            "affiliation": {},
            "email": ""
        },
        {
            "first": "Yejin",
            "middle": [
                "2019"
            ],
            "last": "Choi",
            "suffix": "",
            "affiliation": {},
            "email": ""
        }
    ],
    "year": "",
    "venue": null,
    "identifiers": {},
    "abstract": "Often sentences of correct news are either made biased towards a particular person or a group of persons or parties or maybe distorted to add some sentiment or importance in it. Engaged readers often are not able to extract the inherent meaning of such synthetic sentences. In Bengali, the news contents of the synthetic sentences are presented in such a rich way that it usually becomes difficult to identify the synthetic part of it. We have used machine learning algorithms to classify Bengali news sentences into synthetic and legitimate and then used some rule-based postprocessing on each of these models. Finally, we have developed a voting based combination of these models to build a hybrid model for Bengali synthetic sentence identification. This is a new task and therefore we could not compare it with any existing work in the field. Identification of such types of sentences may be used to improve the performance of identifying fake news and satire news. Thus, identifying molecular level biasness in news articles.",
    "pdf_parse": {
        "paper_id": "2019",
        "_pdf_hash": "",
        "abstract": [
            {
                "text": "Often sentences of correct news are either made biased towards a particular person or a group of persons or parties or maybe distorted to add some sentiment or importance in it. Engaged readers often are not able to extract the inherent meaning of such synthetic sentences. In Bengali, the news contents of the synthetic sentences are presented in such a rich way that it usually becomes difficult to identify the synthetic part of it. We have used machine learning algorithms to classify Bengali news sentences into synthetic and legitimate and then used some rule-based postprocessing on each of these models. Finally, we have developed a voting based combination of these models to build a hybrid model for Bengali synthetic sentence identification. This is a new task and therefore we could not compare it with any existing work in the field. Identification of such types of sentences may be used to improve the performance of identifying fake news and satire news. Thus, identifying molecular level biasness in news articles.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Abstract",
                "sec_num": null
            }
        ],
        "body_text": [
            {
                "text": "The Bengali language is rich in terms of the usage of its words. It is also a relatively free word order language. By changing the order of the same set of words, the author can add some emphasis to some part of the sentence. It is usually observed that the Bengali sentences are frequently distorted like this way. The number of ways English sentences can be distorted is much less than the number of ways a Bengali sentence can be. But all the distorted sentences not necessarily have added biasness or emphasis.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "Some readers take the inherent meaning of the sentences without getting into involved in the biased part of it. They can take out an overview of the text. But often, an engaged reader gets engaged with the writer's views. Sometimes it is not so harmful or it is preferred to be an engaged reader. For example, to get the full flavour of a literary work, the reader has to be engaged. But often it is not desirable. For example, in a piece of political news, it is not recommended to engage a reader without his concern. So, it is essential to notify the reader about synthetic sentences.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "Often sentences of correct news are either made biased towards a particular person or a group of persons or parties or maybe distorted to add some sentiment or importance in it. We refer such types of sentences as synthetic sentences. Engaged readers often are not able to extract the inherent meaning of synthetic sentences. In Bengali, the news contents of the synthetic sentences are presented in such a rich way that it usually becomes difficult to identify the synthetic part of it.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "In this paper, we wish to identify the synthetic sentences in Bengali news automatically. We use the linguistic features in multiple Binary Machine Learning Classifiers to decide whether it is synthetic or legitimate. Then analyzing a confusion matrix, we apply a set of rules. Finally, we combine these models using a voting-based approach. We test this hybrid technique in a Bengali news corpus covering the news in Politics, Sports, Entertainment, and Social domains. The final hybrid technique is able to provide 86% accuracy.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "The rest of the paper is organized as follows. Section 2 illustrates a background study related to synthetic news detection. Section 3 and Section 4 discuss how we have prepared the experimental dataset and model building part. Section 5 shows the results of different steps. Finally, Section 6 presents concluding remarks of the task.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "There are some works in the detection of fake news. They detect fake news in a news corpus Bovet and Makse (2019) ; Batchelor (2017) ; Shu et al. (2017) ; Conroy et al. (2015) that is, misleading news stories which come from non-reputable sources. These papers mainly focus on fake news from four perspectives: the false knowledge, its writing style, its propagation patterns, and the credibility of its creators and spreaders. Rubin et al. (2016) describes three types of fake news in contrast to reporting. These are -serious fabrications (uncovered in mainstream or participant media); large-scale hoaxes; humorous fakes (news satire, parody).",
                "cite_spans": [
                    {
                        "start": 91,
                        "end": 113,
                        "text": "Bovet and Makse (2019)",
                        "ref_id": "BIBREF1"
                    },
                    {
                        "start": 116,
                        "end": 132,
                        "text": "Batchelor (2017)",
                        "ref_id": "BIBREF0"
                    },
                    {
                        "start": 135,
                        "end": 152,
                        "text": "Shu et al. (2017)",
                        "ref_id": "BIBREF8"
                    },
                    {
                        "start": 428,
                        "end": 447,
                        "text": "Rubin et al. (2016)",
                        "ref_id": "BIBREF7"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Related Work",
                "sec_num": "2"
            },
            {
                "text": "Zellers et al. (2019) discussed the threats posed by automatically generated propaganda articles that closely mimics the style of real news. They have designed a language model-based system called Grover for the controllable generation of text from the title of the news. Humans may find this generated text to be more trustworthy compared to the actual news article. Such type of fake news called neural fake news is discriminated best using the generator system itself. Bradshaw and Howard (2017) compared the teams who spread manipulated information, also called disinformation through social media and news across 28 countries including India. These types of fake news are created manually to influence the voters and domestic audiences purposely. Melford and Fagan (2019) designs a Global Disinformation Index (GDI) to combat the disinformation.",
                "cite_spans": [
                    {
                        "start": 472,
                        "end": 498,
                        "text": "Bradshaw and Howard (2017)",
                        "ref_id": "BIBREF2"
                    },
                    {
                        "start": 752,
                        "end": 776,
                        "text": "Melford and Fagan (2019)",
                        "ref_id": "BIBREF5"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Related Work",
                "sec_num": "2"
            },
            {
                "text": "Another essential type of fake news is created by proliferating stylistic bias in the text. P\u00e9rez-Rosas et al. (2017) have used linguistic features in Support Vector Machine (SVM) to detect these fake news in some English newspapers. Rubin et al. (2016) discriminated between synthetic and legitimate news using 5 features namely Absurdity, Humor, Grammar, Negative Affect, and Punctuation.",
                "cite_spans": [
                    {
                        "start": 92,
                        "end": 117,
                        "text": "P\u00e9rez-Rosas et al. (2017)",
                        "ref_id": "BIBREF6"
                    },
                    {
                        "start": 234,
                        "end": 253,
                        "text": "Rubin et al. (2016)",
                        "ref_id": "BIBREF7"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Related Work",
                "sec_num": "2"
            },
            {
                "text": "We evaluate our proposed framework on two datasets, Kaggle Bengali news, and Online Bengali news. For the time being, we are not using the name of the newspapers to avoid the controversy. In total, we have 25K news covering seven different domains, namely Kolkata, State, National, Sports, Entertainment, World, and Travel. Each ",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Dataset Preparation",
                "sec_num": "3"
            },
            {
                "text": "The structure of the news dataset is listed below: 1. Source: Author or publisher of the news article.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "News Content Features",
                "sec_num": "3.1"
            },
            {
                "text": "2. Domain: The domain of the news is defined in this field. In this dataset, we have seven different domains viz, Kolkata, State, National, Sports, Entertainment, World, and Travel.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "News Content Features",
                "sec_num": "3.1"
            },
            {
                "text": "3. Headline: Short title text that aims to catch the attention of readers and describes the main topic of the article 4. Body Text: Main text that elaborates the details of the news story, there is usually a significant claim that shapes the angle of the publisher.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "News Content Features",
                "sec_num": "3.1"
            },
            {
                "text": "Depending on these raw content attributes, different kinds of feature representations can be built to extract discriminative characteristics of synthesis news. The news content we are looking at mostly be linguistic-based features, discussed in the following section.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "News Content Features",
                "sec_num": "3.1"
            },
            {
                "text": "The overall framework of the machine learningbased classifier is divided into three parts: Cleaning of raw text, feature extraction and synthetic news classification. The collected news sentences are annotated manually as a legitimate or synthetic sentence. It is difficult to deal with raw news due to noise. The noisy news includes:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Preprocessing of News Sentences",
                "sec_num": "3.2"
            },
            {
                "text": "\u2022 Keyphrases: -\u09bf\u09a8\u099c\u09b8\u09ac\u09cd \u09aa\u09f0\u09cd\u09bf\u09a4\u09c7\u09ac\u09a6\u09a8, \u0993\u09c7\u09df\u09ac \u09c7\u09a1\u09b8\u09cd\u0995:, \u098f\u0987 \u09bf\u09ac\u09b7\u09c7\u09df \u0985\u09a8 \u09be\u09a8 \u0996\u09ac\u09b0, \u09ac\u09c1 \u09c7\u09b0\u09be, \u09bf\u09a1\u09bf\u099c\u099f\u09be\u09b2 \u09c7\u09a1\u09b8\u09cd\u0995, \u09b8\u09c2 \u09c7\u09a4\u09f0\u09cd\u09b0 \u0996\u09ac\u09b0 Different publication media use the mentioned key phrases which are actually not part of the news. We remove these phrases from the news sentences.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Preprocessing of News Sentences",
                "sec_num": "3.2"
            },
            {
                "text": "\u2022 English Sentences: News contains some English sentences along with Bengali sentences. The following English sentence is highlighted inside a Bengali sports news, e.g.,\"Delhiites get a bite of #ViratKohli quite literally at #MadameTussauds PC Statesman pic.twitter.com/FNLARdIQi6 -Bharat Sharma (sharmabharat45) June 7, 2018\", \"ISIS,\" \"JNU\". We remove such non Bengali sentences from our dataset.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Preprocessing of News Sentences",
                "sec_num": "3.2"
            },
            {
                "text": "\u2022 Stop Words: Stop words are described as the most common words that occur in any corpus of a particular language. At the preprocessing step, we remove stop words such as -'\u098f', '\u098f\u09ac\u0982', '\u0986\u09b0' from the sentences. Here we created a stop word list of 360 words and then these words are removed from the corpus.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Preprocessing of News Sentences",
                "sec_num": "3.2"
            },
            {
                "text": "\u2022 Word Stemming: Word stemming is applied to map the words with different endings to a single one such as \u099a \u09be\u09c7\u09b2\u09c7\u099e\u09cd\u099c\u09b0,\u099a \u09be\u09c7\u09b2\u099e\u09cd\u099c becomes \u099a \u09be\u09c7\u09b2\u099e\u09cd\u099c. Bengali is a very inflectional language for which stemming is required for further processing.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Preprocessing of News Sentences",
                "sec_num": "3.2"
            },
            {
                "text": "\u2022 Other: News contains emoticons, symbols, and pictographs. We remove them by using Unicode.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Preprocessing of News Sentences",
                "sec_num": "3.2"
            },
            {
                "text": "By using the above-mentioned list of phrases, we preprocess the raw news and generate a clean text for further processing.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Preprocessing of News Sentences",
                "sec_num": "3.2"
            },
            {
                "text": "We have annotated Bengali news sentences into two categories: synthetic and legitimate. In this section, we discuss the method we have followed in tagging with some examples. In the cause-effect sentence of Example-3, the cause is a true fact, but the relationship is based on an assumption or probability. There is no rule in the environment to state that one will die after seeing one's six. Therefore this sentence is tagged as synthetic.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Annotation Guideline",
                "sec_num": "3.3"
            },
            {
                "text": "\u2022 Example-4: \u09b8\u09ae\u09cd\u09ad\u09ac\u09a4 \u09ae\u09a8 \u0996\u09be\u09b0\u09be\u09c7\u09aa\u09b0 \u0995\u09be\u09b0\u09c7\u09a8, \u09c7\u09ac\u09be\u09b2\u09cd\u099f \u09a4\u09be\u09b0 \u09c7\u09b8\u09b0\u09be \u09aa\u09be\u09b0\u09ab\u09b0\u09ae \u09be\u09a8\u09cd\u09b8 \u09bf\u09a6\u09c7\u09a4 \u09aa\u09be\u09b0\u09b2\u09a8\u09be\u0964",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Annotation Guideline",
                "sec_num": "3.3"
            },
            {
                "text": "[Probably due to distress, Bolt could not give his best performance.] This is also a cause-effect sentence. In this sentence, the cause is not a true fact as a probability is associated with it. Though, the relation \"if somebody is in distress, then he will not be able to give his best performance\" is an established relation this sentence has synthetic property.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Annotation Guideline",
                "sec_num": "3.3"
            },
            {
                "text": "\u2022 Example-5: The phrases containing synthetic adjectives have a synthetic property like: \u09ae\u09a8 \u09ad\u09be\u0999\u09be \u09c7\u09ac\u09be\u09b2\u09cd\u099f \u09ac\u09b2\u09c7\u09b2\u09a8 [broken heart Bolt told ] We observed that verbs carry the best clue about the synthetic property. Therefore we have created a clue verb list and used it as a feature. But there are some adjectives, adverbs, nouns, which can also be considered a clue. We considered them during annotation.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Annotation Guideline",
                "sec_num": "3.3"
            },
            {
                "text": "\u2022 ",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Annotation Guideline",
                "sec_num": "3.3"
            },
            {
                "text": "After preprocessing of the raw news, the news is tokenized and segmented into sentences level. In this paper, we create a hybrid approach by combining Binary Machine Learning Classifiers using the Voting approach and then postprocessing with a Rule-Based approach. The proposed system is as follows:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Proposed System",
                "sec_num": "4"
            },
            {
                "text": "1. For machine learning, features are generated from the sentences, and after that, we apply supervised machine learning algorithms, namely Support Vector Machine, Naive Bayes, K Nearest Neighbors, Random Forest, Decision Tree, and Logistic Regression.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Proposed System",
                "sec_num": "4"
            },
            {
                "text": "2. According to the results of supervised algorithms, we are creating some rules based on the mismatched outputs.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Proposed System",
                "sec_num": "4"
            },
            {
                "text": "3. Lastly, we are combining the supervised algorithms using a voting approach. We are giving a higher preference for synthetic tagging.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Proposed System",
                "sec_num": "4"
            },
            {
                "text": "If among the six classifiers, 3 classifiers tell the sentence is synthetic and 3 classifiers tell it is legitimate, then we annotate it as synthetic. However, if more than 3 classifiers tell that the sentence is legitimate, then it became legitimate.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Proposed System",
                "sec_num": "4"
            },
            {
                "text": "In this paper, we consider synthetic news classification for independent sentences as each sentence carries some synthetic or legitimate property. Our approach is to use a committee of classifiers, each trained on a set of text features. The entire list of features is presented in this section.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Proposed System",
                "sec_num": "4"
            },
            {
                "text": "Feature Selection of any classification problem takes a crucial part. Each sentence is represented by a feature vector which contains numerical features that represent the occurrence frequency or weight of a feature or binary features (occurrence or non-occurrence of a feature) or a ternary feature. The features are listed below:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Feature Selection in Synthetic Classification",
                "sec_num": "4.1"
            },
            {
                "text": "1. Punctuation: Various punctuation is used to indicate different types of sentiments. The use of punctuation can help the synthetic news detection algorithm to differentiate between funny, entertaining, deceptive, and truthful texts. This feature has two values (binary) -Exclamation (!) and Question Mark (?). This feature is used because, according to our observation the exclamatory sentences and question sentences are more prone to be synthetic. This feature helped us to improve the precision of synthetic sentence identification.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Feature Selection in Synthetic Classification",
                "sec_num": "4.1"
            },
            {
                "text": "2. Named Entities: News tells a story related to a particular incidence. Legitimate news is having the property of some person telling something or making comments. Named Entities are used many times inside news. We have observed that the sentences having a Named Entity mostly become a legitimate sentence. This is because even if an incidence is false or synthetic, but it is told by a particular person we consider it to be true. Consider the example: \"a person x told that he is probably sick\". Here the 'probably' word makes the statement synthetic. But the sentence is legitimate as the person x has actually told it. Thus, use this binary feature (there exist a Named Entity or not) may be helpful to predict synthetic news sentence.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Feature Selection in Synthetic Classification",
                "sec_num": "4.1"
            },
            {
                "text": "To find the named entity in the news sentence, we manually annotate 42k words from 6k news sentences to train CRF++ model. Here, we are using POS tag as a feature of CRF++ to find the presence of named entity in the sentence. To get the POS tag we have used some of the features proposed by Dandapat et al. 2007 which are available with us. Along with the POS tag the Bengali gazetteer list of 1200 names is also used in training the CRF++ model. Then the model is tested with our 530 news examples and we get 94.3% accuracy. We have used this Bengali gazetteer list as a binary feature (presence or non-presence of named entity).",
                "cite_spans": [
                    {
                        "start": 291,
                        "end": 311,
                        "text": "Dandapat et al. 2007",
                        "ref_id": "BIBREF4"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Feature Selection in Synthetic Classification",
                "sec_num": "4.1"
            },
            {
                "text": "3. Domain-specific Clue Verb: We have seen that maximum synthetic sentences have some verbs which carry some indication for the sentence to have synthetic property. Similarly, some verbs are indications of the legitimate property of the sentence. Theses verbs are defined as Clue Verbs. We are attempting to make a list of words for Legitimate sentences and another for synthetic sentences manually. By analyzing a large corpus, we make a list as shown in 1, which is not exhaustive.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Feature Selection in Synthetic Classification",
                "sec_num": "4.1"
            },
            {
                "text": "We have considered this clue verbs as a ternary feature as follows. If there is a Legitimate Clue Verb then it is 1 (we do not need to consider Synthetic Clue Verb); if there is no Legitimate Clue Verb, but there is a Synthetic Clue Verb then it is 2, and if there is neither Legitimate Clue Verb nor Synthetic Clue Verb then it is 3. The value 1 indicates that the sentence is a strong candidate for being legitimate; the value 2 indicates that the sentence is a strong candidate for being synthetic and the value 3 indicates that there is no clue about its property. According to our observation, this feature is the most effecting feature in the classification task.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Feature Selection in Synthetic Classification",
                "sec_num": "4.1"
            },
            {
                "text": "Term Frequency -Inverse Document Frequency (TF-IDF) of a term is used to denote its importance. TF(w,d) denotes the raw count of the word (w) in a news document (d) and IDF(w,D) is a measure of how much information the word (w) provides, i.e., if it is usual or rare across all news documents (D). Finally, TF-IDF is defined as follows.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "TF-IDF:",
                "sec_num": "4."
            },
            {
                "text": "EQUATION",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [
                    {
                        "start": 0,
                        "end": 8,
                        "text": "EQUATION",
                        "ref_id": "EQREF",
                        "raw_str": "tf _idf (w, d, D) = tf (w, d) \u00d7 idf (w, D)",
                        "eq_num": "(1)"
                    }
                ],
                "section": "TF-IDF:",
                "sec_num": "4."
            },
            {
                "text": "Thus, TF-IDF is used to determine the importance of words in news domain. We have combined the TF-IDF of words of the input sentence to calculate the importance of sentence in news domain. Considering that the synthetic sentences carry more importance we have used it as a feature in our task.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "TF-IDF:",
                "sec_num": "4."
            },
            {
                "text": "Synthetic news sentence classification may be done at the document level, sentence level, and phrase level. We are considering the news article is based on an actual fact. Some of its sentences are synthesized by the author to attract the engaged reader. Our objective is to identify those synthetic sentences. So, in this paper, sentence-level classification is considered where an independent sentence is classified as synthetic sentences and legitimate sentences.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Machine Learning-Based Classification",
                "sec_num": "4.2"
            },
            {
                "text": "A supervised binary classifier algorithm may be used to identify the synthetic sentences. Several supervised machine learning techniques have been examined in the paper to classify the sentences into classes. Those are Support Vector Machine, Naive Bayes, K Nearest Neighbors, Random Forest, Decision Tree, and Logistic Regression (LR). We have considered the Punctuation feature, Named Entity feature, clue verb, and TF-IDF feature, as discussed in Section 4.1.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Machine Learning-Based Classification",
                "sec_num": "4.2"
            },
            {
                "text": "The Confusion Matrix is one of the most intuitive metrics used for finding the correctness and accuracy of the model. The Confusion Matrix in itself is not a performance measure as such, but almost all of the performance metrics are based on the Confusion Matrix and the numbers inside it. The confusion matrix is a table with two dimensions (\"Actual\" and \"Predicted\"), and sets of \"classes\" in both dimensions.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Machine Learning-Based Classification",
                "sec_num": "4.2"
            },
            {
                "text": "The following terms are associated with a confusion matrix.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Machine Learning-Based Classification",
                "sec_num": "4.2"
            },
            {
                "text": "\u2022 True Positive (TP): when predicted synthetic sentences pieces are actually annotated as a synthetic sentence;",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Machine Learning-Based Classification",
                "sec_num": "4.2"
            },
            {
                "text": "\u2022 True Negative (TN): when predicted legitimate sentences pieces are actually annotated as true sentences; \u2022 False Negative (FN): when predicted legitimate sentences pieces are actually annotated as synthetic sentences;",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Machine Learning-Based Classification",
                "sec_num": "4.2"
            },
            {
                "text": "\u2022 False Positive (FP): when predicted synthetic sentences, pieces are actually annotated as legitimate sentences.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Machine Learning-Based Classification",
                "sec_num": "4.2"
            },
            {
                "text": "We have created a separate set of 106 sentences to create the Confusion Matrix. The Confusion Matrices of all the Binary Classifier techniques for these sentences are shown in Fig. 2 .",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 176,
                        "end": 182,
                        "text": "Fig. 2",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Machine Learning-Based Classification",
                "sec_num": "4.2"
            },
            {
                "text": "After analyzing the errors in the confusion matrix of the Binary Classifier techniques, as shown in Fig. 2 , we have formulated an initial set of rules. These rules are used in the postprocessing step to correct some of the errors. The rules we formulated are discussed below.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 100,
                        "end": 106,
                        "text": "Fig. 2",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Rule Based Postprocessing",
                "sec_num": "4.3"
            },
            {
                "text": "1. If the topic of the news sentence is clubbed with old news of different topic, then it is considered as synthetic. We have prepared an initial list of abstract reasons. In this sentence, the reason \"infighting\" is not concrete. Therefore it is considered synthetic.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Rule Based Postprocessing",
                "sec_num": "4.3"
            },
            {
                "text": "3. If there is an incomplete list of names, matters, topics, or any other thing in the sentence, then it is considered as synthetic. Consider the following example. ",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Rule Based Postprocessing",
                "sec_num": "4.3"
            },
            {
                "text": "Various evaluation metrics have been used to evaluate the performance of machine learning models. We want to test the performances of our models in terms of Accuracy, Precision, Recall, and F1-Score. These metrics are commonly used to evaluate the machine learning models and enable us to evaluate the performance of a classifier from different perspectives. The results of the k-fold cross-validations for each of our hybrid models are shown in section 5.1. Then we have applied the voting approach to combine the models and then the rules. The final accuracy of this hybrid system is discussed in section 5.2.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Experimental Result",
                "sec_num": "5"
            },
            {
                "text": "Firstly, we have tested the six binary classifiers namely Logistic Regression, K Nearest Neighbors, Support Vector Machine, Naive Bayes, Decision Tree, and Random Forest. We have used 530 sentences annotated as legitimate or synthetic. These sentences are folded into Training and Test data in the proportion of 80:20. We have used k-fold cross-validation (k = 5) and calculated the above metrics in each of the kfolds for each model. Then, we have manually applied the rules on all the six prediction models. We prepared a comparison chart of the final performances of these six individual hybrid systems which is given in Fig. 3 .",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 624,
                        "end": 630,
                        "text": "Fig. 3",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Classification Performance of Individual Hybrid Machine Learning Models",
                "sec_num": "5.1"
            },
            {
                "text": "In Fig. 3 , the dark gray colour bars indicate the k-fold cross-validation results of the binary clas- Table 2 : Performances of different approaches for Synthetic Sentence classification Figure 3 : Classifier Result sification. The light gray colour bars indicate the result we got after applying the rules on the best output of the corresponding technique. The result shows that the rules improved the Support Vector Machine and LR based techniques most. The Logistic Regression based technique combined with the rules gave the highest accuracy up to this stage. We are getting highest around 82% accuracy by applying rules on the Logistic Regression based model.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 3,
                        "end": 9,
                        "text": "Fig. 3",
                        "ref_id": null
                    },
                    {
                        "start": 103,
                        "end": 110,
                        "text": "Table 2",
                        "ref_id": null
                    },
                    {
                        "start": 188,
                        "end": 196,
                        "text": "Figure 3",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Classification Performance of Individual Hybrid Machine Learning Models",
                "sec_num": "5.1"
            },
            {
                "text": "Finally, we have used a voting based combination of these six machine learning classifiers. If a sentence is tagged as synthetic by 3 or more classifiers then we consider it to be synthetic. Otherwise, it is considered to be legitimate. Then, we have applied the rules on the combined classifier. The final result of this hybrid system is shown in Table 2 .",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 348,
                        "end": 355,
                        "text": "Table 2",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Classification Performance of Combined Hybrid Machine Learning Model",
                "sec_num": "5.2"
            },
            {
                "text": "In this paper, we defined and compared synthetic and legitimate sentences and highlighted many interesting differences between these two categories.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Conclusion and Future Scope",
                "sec_num": "6"
            },
            {
                "text": "We then utilized these differences as features to detect synthetic sentences. We have proposed a hybrid approach that can detect synthetic news. To the best of our knowledge, our work is the first attempt to detect synthetic news at the Bengali news sentence level. In the future, we want to extend it to use semantic features in the Machine Learning model and calculate the degree of synthetic property in the synthetic sentences. Then we want to compare the renowned Bengali newspapers in terms of the usage of different types of synthetic sentences.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Conclusion and Future Scope",
                "sec_num": "6"
            }
        ],
        "back_matter": [],
        "bib_entries": {
            "BIBREF0": {
                "ref_id": "b0",
                "title": "Getting out the truth: the role of libraries in the fight against fake news. Reference services review",
                "authors": [
                    {
                        "first": "Oliver",
                        "middle": [],
                        "last": "Batchelor",
                        "suffix": ""
                    }
                ],
                "year": 2017,
                "venue": "",
                "volume": "45",
                "issue": "",
                "pages": "143--148",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Oliver Batchelor. 2017. Getting out the truth: the role of libraries in the fight against fake news. Reference services review, 45(2):143-148.",
                "links": null
            },
            "BIBREF1": {
                "ref_id": "b1",
                "title": "Influence of fake news in twitter during the 2016 us presidential election",
                "authors": [
                    {
                        "first": "Alexandre",
                        "middle": [],
                        "last": "Bovet",
                        "suffix": ""
                    },
                    {
                        "first": "A",
                        "middle": [],
                        "last": "Hern\u00e1n",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Makse",
                        "suffix": ""
                    }
                ],
                "year": 2019,
                "venue": "Nature communications",
                "volume": "10",
                "issue": "1",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Alexandre Bovet and Hern\u00e1n A Makse. 2019. Influ- ence of fake news in twitter during the 2016 us pres- idential election. Nature communications, 10(1):7.",
                "links": null
            },
            "BIBREF2": {
                "ref_id": "b2",
                "title": "Troops, trolls and troublemakers: A global inventory of organized social media manipulation",
                "authors": [
                    {
                        "first": "Samantha",
                        "middle": [],
                        "last": "Bradshaw",
                        "suffix": ""
                    },
                    {
                        "first": "Philip",
                        "middle": [],
                        "last": "Howard",
                        "suffix": ""
                    }
                ],
                "year": 2017,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Samantha Bradshaw and Philip Howard. 2017. Troops, trolls and troublemakers: A global inventory of orga- nized social media manipulation. Technical report, Oxford Internet Institute.",
                "links": null
            },
            "BIBREF3": {
                "ref_id": "b3",
                "title": "Automatic deception detection: Methods for finding fake news",
                "authors": [
                    {
                        "first": "J",
                        "middle": [],
                        "last": "Niall",
                        "suffix": ""
                    },
                    {
                        "first": "Victoria",
                        "middle": [
                            "L"
                        ],
                        "last": "Conroy",
                        "suffix": ""
                    },
                    {
                        "first": "Yimin",
                        "middle": [],
                        "last": "Rubin",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Chen",
                        "suffix": ""
                    }
                ],
                "year": 2015,
                "venue": "Proceedings of the Association for Information Science and Technology",
                "volume": "52",
                "issue": "",
                "pages": "1--4",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Niall J Conroy, Victoria L Rubin, and Yimin Chen. 2015. Automatic deception detection: Methods for finding fake news. Proceedings of the Association for Information Science and Technology, 52(1):1-4.",
                "links": null
            },
            "BIBREF4": {
                "ref_id": "b4",
                "title": "Automatic part-of-speech tagging for bengali: An approach for morphologically rich languages in a poor resource scenario",
                "authors": [
                    {
                        "first": "Sandipan",
                        "middle": [],
                        "last": "Dandapat",
                        "suffix": ""
                    },
                    {
                        "first": "Sudeshna",
                        "middle": [],
                        "last": "Sarkar",
                        "suffix": ""
                    },
                    {
                        "first": "Anupam",
                        "middle": [],
                        "last": "Basu",
                        "suffix": ""
                    }
                ],
                "year": 2007,
                "venue": "Proceedings of the 45th Annual Meeting of the ACL on Interactive Poster and Demonstration Sessions",
                "volume": "",
                "issue": "",
                "pages": "221--224",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Sandipan Dandapat, Sudeshna Sarkar, and Anupam Basu. 2007. Automatic part-of-speech tagging for bengali: An approach for morphologically rich lan- guages in a poor resource scenario. In Proceedings of the 45th Annual Meeting of the ACL on Interactive Poster and Demonstration Sessions, pages 221-224. Association for Computational Linguistics.",
                "links": null
            },
            "BIBREF5": {
                "ref_id": "b5",
                "title": "Cutting the funding of disinformation: The ad-tech solution",
                "authors": [
                    {
                        "first": "Clare",
                        "middle": [],
                        "last": "Melford",
                        "suffix": ""
                    },
                    {
                        "first": "Craig",
                        "middle": [],
                        "last": "Fagan",
                        "suffix": ""
                    }
                ],
                "year": 2019,
                "venue": "The Global Disinformation Index",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Clare Melford and Craig Fagan. 2019. Cutting the funding of disinformation: The ad-tech solution. Technical report, The Global Disinformation Index.",
                "links": null
            },
            "BIBREF6": {
                "ref_id": "b6",
                "title": "Automatic detection of fake news",
                "authors": [
                    {
                        "first": "Ver\u00f3nica",
                        "middle": [],
                        "last": "P\u00e9rez-Rosas",
                        "suffix": ""
                    },
                    {
                        "first": "Bennett",
                        "middle": [],
                        "last": "Kleinberg",
                        "suffix": ""
                    },
                    {
                        "first": "Alexandra",
                        "middle": [],
                        "last": "Lefevre",
                        "suffix": ""
                    },
                    {
                        "first": "Rada",
                        "middle": [],
                        "last": "Mihalcea",
                        "suffix": ""
                    }
                ],
                "year": 2017,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "arXiv": [
                        "arXiv:1708.07104"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Ver\u00f3nica P\u00e9rez-Rosas, Bennett Kleinberg, Alexan- dra Lefevre, and Rada Mihalcea. 2017. Auto- matic detection of fake news. arXiv preprint arXiv:1708.07104.",
                "links": null
            },
            "BIBREF7": {
                "ref_id": "b7",
                "title": "Fake news or truth? using satirical cues to detect potentially misleading news",
                "authors": [
                    {
                        "first": "Victoria",
                        "middle": [],
                        "last": "Rubin",
                        "suffix": ""
                    },
                    {
                        "first": "Niall",
                        "middle": [],
                        "last": "Conroy",
                        "suffix": ""
                    },
                    {
                        "first": "Yimin",
                        "middle": [],
                        "last": "Chen",
                        "suffix": ""
                    },
                    {
                        "first": "Sarah",
                        "middle": [],
                        "last": "Cornwell",
                        "suffix": ""
                    }
                ],
                "year": 2016,
                "venue": "Proceedings of the Second Workshop on Computational Approaches to Deception Detection",
                "volume": "",
                "issue": "",
                "pages": "7--17",
                "other_ids": {
                    "DOI": [
                        "10.18653/v1/W16-0802"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Victoria Rubin, Niall Conroy, Yimin Chen, and Sarah Cornwell. 2016. Fake news or truth? using satir- ical cues to detect potentially misleading news. In Proceedings of the Second Workshop on Computa- tional Approaches to Deception Detection, pages 7- 17, San Diego, California. Association for Compu- tational Linguistics.",
                "links": null
            },
            "BIBREF8": {
                "ref_id": "b8",
                "title": "Exploiting tri-relationship for fake news detection",
                "authors": [
                    {
                        "first": "Kai",
                        "middle": [],
                        "last": "Shu",
                        "suffix": ""
                    },
                    {
                        "first": "Suhang",
                        "middle": [],
                        "last": "Wang",
                        "suffix": ""
                    },
                    {
                        "first": "Huan",
                        "middle": [],
                        "last": "Liu",
                        "suffix": ""
                    }
                ],
                "year": 2017,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "arXiv": [
                        "arXiv:1712.07709"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Kai Shu, Suhang Wang, and Huan Liu. 2017. Exploit- ing tri-relationship for fake news detection. arXiv preprint arXiv:1712.07709.",
                "links": null
            }
        },
        "ref_entries": {
            "FIGREF0": {
                "num": null,
                "uris": null,
                "text": "Bengali News Dataset Distribution in Kolkata, State, National, Sports, Entertainment, World, and Travel Domains news contains on an average of 15 sentences. The distribution of the Bengali news dataset in the seven domains is shown inFig. 1",
                "type_str": "figure"
            },
            "FIGREF1": {
                "num": null,
                "uris": null,
                "text": "Figure 2: Confusion Matrix after Applying 6 Binary Classifiers on 106 Sentences[ X axis denotes predicted value and Y axis denotes actual value and both cases 0 indicates legitimate sentence and 1 indicates Synthetic Sentence ]",
                "type_str": "figure"
            },
            "TABREF0": {
                "num": null,
                "html": null,
                "content": "<table/>",
                "type_str": "table",
                "text": "Example-1: \u098f\u0987 \u09b8\u09ae\u09cd\u09aa\u09f0\u09cd\u09a6\u09be\u09c7\u09df\u09b0 \u098f\u0995 \u09b8\u09a6\u09c7\u09b8 \u09b0 \u09ae\u09c7\u09a4, \u09e8\u09e6\u09e7\u09eb \u09b8\u09be\u09c7\u09b2 \u0986\u09c7\u09ae\u09bf\u09b0\u0995\u09be\u09c7\u09a4 \u098f\u09ac\u0982 \u09e8\u09e6\u09e7\u09ed \u09b8\u09be\u09c7\u09b2 \u0987\u0982\u09b2 \u09be\u09c7\u09a8\u09cd\u09a1 \u09b8\u09ad\u09be \u0995\u09c7\u09b0\u09bf\u099b\u09c7\u09b2\u09a8 \u09aa\u09f0\u09cd\u09a7\u09be\u09a8\u09ae\u09a8\u09cd\u09a4\u09f0\u09cd\u09c0\u0964 [According to a member of this community, the Prime Minister had a meeting in the United States in 2015 and England in 2017. ]In this sentence, it is claimed that the statement is taken from somebody, but the name is not mentioned explicitly. This is why, we consider such types of sentences as synthetic. If the name of the claimer is added, then it is converted to legitimate."
            },
            "TABREF2": {
                "num": null,
                "html": null,
                "content": "<table><tr><td>Label</td><td>Clue Verb</td></tr><tr><td>Legitimate</td><td>\u09ac\u09b2\u09c7\u099b\u09a8 [\u0995\u09c7\u09b0\u09c7\u099b[The question has started to arise], \u0986\u09b8\u09cd\u09a5\u09be \u09c7\u09a8\u0987[Not confident], \u09ae\u09c7\u09a8 \u0995\u09b0\u09be \u09b9\u09c7\u099a\u09cd\u099b[It seems ], \u099a \u09be\u09c7\u09b2\u099e\u09cd\u099c \u09bf\u09a8\u09c7\u09df\u09c7\u099b\u09a8 [Have taken up the challenge ], \u099a \u09be\u09c7\u09b2\u099e\u09cd\u099c \u09c7\u099b\u09be\u09c7\u09dc\u09a8[Throw the challenge], \u099a \u09be\u09c7\u09b2\u099e\u09cd\u099c \u09c7\u09a8\u0993\u09df\u09be [Take up the challenge], \u099a \u09be\u09c7\u09b2\u099e\u09cd\u099c \u09c7\u099b\u09be\u09dc\u09be[Throw the challenge], \u099a \u09be\u09c7\u09b2\u099e\u09cd\u099c[challenge-H ],\u09ae\u09c7\u09a8 \u0995\u09b0\u09be \u09b9\u09c7\u099a\u09cd\u099b[It seems], \u09aa\u09f0\u09cd\u09ae\u09be\u09bf\u09a3\u09a4 \u09b9\u09c7\u09ac[Will prove], \u09a8\u09c0\u09bf\u09a4 \u09bf\u09a8\u09c7\u09df\u09c7\u099b[Policy taken], \u09b8\u09bf\u0995\u09f0\u09cd\u09df \u09ad\u09be\u09c7\u09ac \u09aa\u09c7\u09a5 \u09a8\u09be\u09ae\u09c7\u09a4 \u09c7\u09a6\u0996\u09be \u09bf\u0997\u09c7\u09df\u09c7\u099b[Actively shown on the way down], \u09a6\u09be\u09bf\u09ac &lt;Null verb&gt;[Claim-H], \u0996\u09ac\u09b0 \u099a\u09be\u0989\u09b0 \u09b9\u09c7\u09df \u09af\u09be\u09df[The news goes sour], \u0986\u09a4\u09c7\u0999\u09cd\u0995\u09b0 \u099b\u09be\u09aa &lt;Null Verb&gt;[The impression of terror], \u0986\u09a4\u09c7\u0999\u09cd\u0995\u09b0 \u099b\u09be\u09aa \u09c7\u09a8\u0987[No sign of panic], \u09a4\u09a6\u09be\u09b0\u09bf\u0995 \u0995\u09c7\u09b0\u09a8[Take care], \u09c7\u09ab\u09b0 \u09b8\u09a4\u0995\u09b0\u09cd \u0995\u09c7\u09b0 \u09c7\u09a6\u09a8[Warns again], \u09a6\u09be\u09bf\u09ac \u09a4\u09c1 \u09c7\u09b2\u09bf\u099b\u09c7\u09b2\u09a8[Claimed-H], \u09a4\u09ac\u09c1 \u09b8\u09ac\u09cd\u09bf\u09b8\u09cd\u09a4 \u09bf\u099b\u09b2[Yet there was relief], \u09c7\u09a6\u0996\u09c7\u099b\u09a8 \u09b8\u09cd\u09a5\u09be\u09a8\u09c0\u09df\u09b0\u09be[The locals are watching], \u0986\u09b8\u09cd\u09a5\u09be \u09c7\u09a8\u0987[Not con-fident], \u0986\u09b6\u09ac\u09cd\u09be\u09b8 \u09bf\u09a6\u09c7\u09df\u09c7\u099b\u09a8[Assured-H], \u0986\u09b6\u0999\u09cd\u0995\u09be[Fear], \u0986\u09b6\u0999\u09cd\u0995\u09be \u0995\u09c7\u09b0\u09c7\u099b\u09a8[Have feared], \u0986\u09b6\u0999\u09cd\u0995\u09be \u0995\u09b0\u09be[Do not be afraid], \u0986\u09b6\u0999\u09cd\u0995\u09be \u0995\u09b0\u09b2[Apprehensive], \u09b0\u09c7\u09df\u09c7\u099b \u09ac\u09c7\u09b2[Say there is], \u09c7\u09a6\u0993\u09df\u09be\u09b0 \u0985\u09bf\u09ad\u09c7\u09af\u09be\u0997[The charge to give], \u099a\u09be\u0999\u09cd\u0997\u09be \u09b9\u09c7\u09df \u0993\u09c7\u09a0[Became stronger], \u0985\u09bf\u09ad\u09c7\u09af\u09be\u0997 \u09a4\u09c1 \u09b2\u09c7\u09b2\u09a8[Complain-H], \u0985\u09bf\u09ad\u09c7\u09af\u09be\u0997 \u0995\u09b0\u09b2[Complained-H], \u0985\u09bf\u09ad\u09c7\u09af\u09be\u0997 \u0995\u09b0\u09be \u09b9\u09c7\u09df\u09c7\u099b[The complaint was made], \u0985\u09bf\u09ad\u09c7\u09af\u09be\u0997[Complain-H], \u0995\u09dc\u09be \u09ac\u09be\u09a4\u09b0\u09cd \u09be[Strong message], \u09b0\u09c1\u09c7\u0996 \u09a6\u09be\u0981 \u09bf\u09dc\u09c7\u09df\u09bf\u099b\u09c7\u09b2\u09a8 [Standing in the stands], \u09b0\u09c1\u09c7\u0996 \u09a6\u09be\u09dc\u09be\u09a8-H[Stand up], \u09b0\u09c1\u09c7\u0996 \u09a6\u09be\u09dc\u09be\u09c7\u09b2\u09a8-H[Stand up], \u09c7\u09a4\u09be\u09c7\u09aa\u09b0 \u09ae\u09c1 \u09c7\u0996-H[Under the cannon], \u09ae\u09be\u09c7\u09a0 \u09a8\u09be\u09ae\u09be\u09c7\u099a\u09cd\u099b\u09a8-H[Getting down on the field], \u09b8\u09ac\u09cd\u09c0\u0995\u09c3 \u09bf\u09a4 \u09bf\u09a6\u09c7\u09df\u09c7\u099b\u09a8[Recognized], \u09b8\u09ac\u09cd\u09c0\u0995\u09c3 \u09bf\u09a4 \u09c7\u09a6\u0993\u09df\u09be[recognition-H], \u09b8\u09ac\u09cd\u09c0\u0995\u09c3 \u09bf\u09a4 \u09bf\u09ae\u09b2\u09c7\u09b2[Acceptance-H], \u09bf\u09ac\u099a\u09be\u09c7\u09b0\u09b0 \u09ae\u09c1 \u09c7\u0996\u09be\u09ae\u09c1 \u09bf\u0996[Facing trial], \u0985\u09bf\u09ad\u09c7\u09af\u09be\u0997 \u0989\u09a0\u09b2[The complaint arose], \u0985\u09bf\u09ad\u09c7\u09af\u09be\u0997 \u0989\u09a0\u09be[Complaints-H arise], \u09ac\u09c7\u09b2 \u0985\u09bf\u09ad\u09c7\u09af\u09be\u0997[Complain-H], \u098f\u0995\u09be\u0982\u09c7\u09b6\u09b0 \u09a6\u09be\u09bf\u09ac \u099b\u09be\u09aa &lt;Null Verb&gt;[Part claim impression], \u098f\u0995\u09be\u0982\u09c7\u09b6\u09b0 \u09ae\u09a4[Like a part], \u0989\u09c7\u09b8\u09cd\u0995 \u09bf\u09a6\u09b2[Instigated-H], \u0995\u09c0\u09c7\u09b8\u09b0 \u0987\u09bf\u0999\u09cd\u0997\u09a4[What a hint], \u099c\u09be\u09a8\u09be \u09bf\u0997\u09c7\u09df\u09c7\u099b[Got it]</td></tr></table>",
                "type_str": "table",
                "text": "Says-H], \u099c\u09be\u09bf\u09a8\u09c7\u09df\u09c7\u099b\u09a8 [Said-H],\u099c\u09be\u09a8\u09be\u09a8 [Said-H ], \u09ac\u09b2\u09c7\u09b2\u09a8 [Said-H],\u0998\u09cb\u09b7\u09a3\u09be\u09ae\u09a4\u09cb [Declaration], \u0995\u09a5\u09be \u09ac\u09c7\u09b2\u09a8[Speak-H]\u0995\u09a5\u09be \u09ac\u09b2\u09c7\u09b2\u09a8 [Speak-H], \u099c\u09be\u09a8\u09be\u09a8[Tell me], \u099c\u09be\u09a8\u09be\u09c7\u09b2\u09a8[Told], \u09c7\u0998\u09be\u09b7\u09a3\u09be \u0995\u09b0\u09c7\u09b2\u09a8[Announced] Synthetic \u0996\u09bf\u09a4\u09c7\u09df \u09c7\u09a6\u0996\u09be[Check it out], \u0996\u09bf\u09a4\u09c7\u09df \u09c7\u09a6\u09c7\u0996\u09a8[Check it out],\u0996\u09a4\u09bf\u09af\u09bc \u09c7 \u09c7\u09a6\u0996\u09be \u09b9\u09df[Is checked ], \u09aa\u09f0\u09cd\u09b6\u09cd\u09a8 \u0995\u09c7\u09b0\u09c7\u099b\u09a8 [Asked-H], \u09aa\u09f0\u09cd\u09b6\u09cd\u09a8 \u0995\u09c7\u09b0 [Asked-H], \u09aa\u09f0\u09cd\u09b6\u09cd\u09a8 \u0995\u09b0\u09be \u09b9\u09c7\u09df\u09c7\u099b[Asked-H], \u09aa\u09f0\u09cd\u09b6\u09cd\u09a8 \u0995\u09b0\u09be[Asked-H], \u09aa\u09f0\u09cd\u09b6\u09cd\u09a8 \u09b9\u09c7\u09df\u09c7\u099b [The question has been], \u09aa\u09f0\u09cd\u09b6\u09cd\u09a8 \u0989\u09c7\u09a0\u09c7\u099b [The question arises], \u0995\u09a5\u09be \u09bf\u099b\u09b2 [There was talk], \u0995\u09a5\u09be \u09c7\u09a6\u0993\u09df\u09be[Promise-H], \u0995\u09a5\u09be \u09bf\u09a6\u09c7\u09df\u09bf\u099b\u09b2[Promised-H], \u0995\u09a5\u09be \u09bf\u09a6\u09c7\u09df\u09bf\u099b\u09c7\u09b2\u09a8[Promised-H],\u0995\u09a5\u09be \u09bf\u09a6\u09c7\u09df \u09b0\u09be\u09c7\u0996\u09a8\u09bf\u09a8[Didn't talk], \u09aa\u09f0\u09cd\u09b6\u09cd\u09a8 \u0989\u09a0\u09c7\u09a4 \u09b6\u09c1\u09b0\u09c1"
            },
            "TABREF3": {
                "num": null,
                "html": null,
                "content": "<table/>",
                "type_str": "table",
                "text": ""
            }
        }
    }
}