sidneibarbieri commited on
Commit
784d8b8
·
verified ·
1 Parent(s): 4b887fe

Demonstration re-recorded with v1.12.0: dark theme and validated chart palette

Browse files
assets/demo/topvenues-demo-v1.12.0.en.vtt ADDED
@@ -0,0 +1,560 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ WEBVTT
2
+
3
+ 1
4
+ 00:00:00.000 --> 00:00:05.722
5
+ Every systematic review rests on a number
6
+ that almost never gets written down.
7
+
8
+ 2
9
+ 00:00:05.762 --> 00:00:06.904
10
+ The denominator.
11
+
12
+ 3
13
+ 00:00:06.944 --> 00:00:11.376
14
+ The set of papers that were
15
+ eligible before screening began.
16
+
17
+ 4
18
+ 00:00:11.976 --> 00:00:14.687
19
+ In practice that set is rebuilt by hand,
20
+
21
+ 5
22
+ 00:00:14.727 --> 00:00:17.576
23
+ from portals and interfaces whose coverage
24
+
25
+ 6
26
+ 00:00:17.616 --> 00:00:20.465
27
+ and query semantics change underneath you.
28
+
29
+ 7
30
+ 00:00:20.505 --> 00:00:25.320
31
+ Two people follow the same protocol
32
+ and end up with different corpora.
33
+
34
+ 8
35
+ 00:00:25.920 --> 00:00:28.460
36
+ TopVenues makes that
37
+ denominator an object.
38
+
39
+ 9
40
+ 00:00:28.500 --> 00:00:30.740
41
+ Declared, frozen, hashed, and citable.
42
+
43
+ 10
44
+ 00:00:30.780 --> 00:00:33.240
45
+ This is what working with one looks like.
46
+
47
+ 11
48
+ 00:00:33.840 --> 00:00:36.561
49
+ Installation is a clone and one script.
50
+
51
+ 12
52
+ 00:00:36.601 --> 00:00:40.170
53
+ Nothing is fetched from a
54
+ publisher at review time,
55
+
56
+ 13
57
+ 00:00:40.210 --> 00:00:43.608
58
+ and nothing depends on
59
+ a service staying online.
60
+
61
+ 14
62
+ 00:00:44.208 --> 00:00:49.686
63
+ The script builds an isolated environment
64
+ from pinned, hash-checked dependencies,
65
+
66
+ 15
67
+ 00:00:49.726 --> 00:00:53.432
68
+ then materializes the corpus
69
+ from a compressed snapshot
70
+
71
+ 16
72
+ 00:00:53.472 --> 00:00:55.584
73
+ that ships with the repository.
74
+
75
+ 17
76
+ 00:00:56.184 --> 00:00:58.685
77
+ Then it verifies rather than asserts.
78
+
79
+ 18
80
+ 00:00:58.725 --> 00:01:02.463
81
+ The snapshot's SHA-256 is
82
+ checked against the manifest,
83
+
84
+ 19
85
+ 00:01:02.503 --> 00:01:07.064
86
+ the record counts are checked
87
+ against the numbers the paper prints,
88
+
89
+ 20
90
+ 00:01:07.104 --> 00:01:09.096
91
+ and the full test suite runs.
92
+
93
+ 21
94
+ 00:01:09.696 --> 00:01:13.094
95
+ Twenty thousand three
96
+ hundred and five records.
97
+
98
+ 22
99
+ 00:01:13.134 --> 00:01:18.727
100
+ Seventeen thousand four hundred and
101
+ ninety-one abstracts. Every test passing.
102
+
103
+ 23
104
+ 00:01:18.767 --> 00:01:23.376
105
+ A reviewer reaches this state
106
+ without asking anyone for access.
107
+
108
+ 24
109
+ 00:01:23.976 --> 00:01:26.909
110
+ The interface opens on
111
+ what you are holding.
112
+
113
+ 25
114
+ 00:01:26.949 --> 00:01:32.584
115
+ Fourteen thousand eight hundred and fifty-nine
116
+ papers across twenty declared venues,
117
+
118
+ 26
119
+ 00:01:32.624 --> 00:01:38.232
120
+ ninety-four percent carrying a searchable
121
+ abstract, and every record ready to cite.
122
+
123
+ 27
124
+ 00:01:38.832 --> 00:01:40.870
125
+ Nothing here is a live query.
126
+
127
+ 28
128
+ 00:01:40.910 --> 00:01:44.166
129
+ The venue list is a
130
+ declared scientific scope,
131
+
132
+ 29
133
+ 00:01:44.206 --> 00:01:48.648
134
+ and changing it is a release
135
+ decision, not a filter you slide.
136
+
137
+ 30
138
+ 00:01:49.248 --> 00:01:51.721
139
+ Start with an ordinary question.
140
+
141
+ 31
142
+ 00:01:51.761 --> 00:01:57.140
143
+ Ranked search orders title, abstract
144
+ and author matches by relevance,
145
+
146
+ 32
147
+ 00:01:57.180 --> 00:02:00.792
148
+ over the whole frozen
149
+ corpus, in milliseconds.
150
+
151
+ 33
152
+ 00:02:01.392 --> 00:02:07.231
153
+ Narrow it the way a protocol narrows: by
154
+ venue tier, by year, by author position.
155
+
156
+ 34
157
+ 00:02:07.271 --> 00:02:13.368
158
+ The count updates, and that count is the
159
+ number you will report as your denominator.
160
+
161
+ 35
162
+ 00:02:13.968 --> 00:02:17.221
163
+ Every row opens into the record behind it.
164
+
165
+ 36
166
+ 00:02:17.261 --> 00:02:23.808
167
+ The abstract that was matched, the venue,
168
+ the tier, the DOI, and the publisher link.
169
+
170
+ 37
171
+ 00:02:23.848 --> 00:02:27.768
172
+ You are reading the
173
+ evidence, not a summary of it.
174
+
175
+ 38
176
+ 00:02:28.368 --> 00:02:29.492
177
+ And it leaves.
178
+
179
+ 39
180
+ 00:02:29.532 --> 00:02:35.644
181
+ CSV for a screening sheet, JSON for
182
+ a pipeline, BibTeX for the manuscript,
183
+
184
+ 40
185
+ 00:02:35.684 --> 00:02:41.088
186
+ taken straight from DBLP's canonical
187
+ entries rather than retyped.
188
+
189
+ 41
190
+ 00:02:41.688 --> 00:02:45.687
191
+ Insights is where the
192
+ corpus stops being a list.
193
+
194
+ 42
195
+ 00:02:45.727 --> 00:02:50.568
196
+ It reads in four passes: what
197
+ it holds, how a topic moves,
198
+
199
+ 43
200
+ 00:02:50.608 --> 00:02:55.488
201
+ who produces that work, and
202
+ where it cannot answer at all.
203
+
204
+ 44
205
+ 00:02:56.088 --> 00:02:57.362
206
+ Composition first.
207
+
208
+ 45
209
+ 00:02:57.402 --> 00:03:00.939
210
+ Papers per venue, per
211
+ year, and per record class.
212
+
213
+ 46
214
+ 00:03:00.979 --> 00:03:04.880
215
+ Full articles are seventy-eight
216
+ percent of the corpus;
217
+
218
+ 47
219
+ 00:03:04.920 --> 00:03:10.176
220
+ the rare classes are the ones a
221
+ protocol has to decide about explicitly.
222
+
223
+ 48
224
+ 00:03:10.776 --> 00:03:12.751
225
+ Every aggregate is a door.
226
+
227
+ 49
228
+ 00:03:12.791 --> 00:03:18.097
229
+ Click a bar and you are in Search,
230
+ filtered to exactly those records.
231
+
232
+ 50
233
+ 00:03:18.137 --> 00:03:21.624
234
+ No chart here is a
235
+ picture you have to trust.
236
+
237
+ 51
238
+ 00:03:22.224 --> 00:03:25.651
239
+ Now the pass that changes
240
+ how you read a field.
241
+
242
+ 52
243
+ 00:03:25.691 --> 00:03:29.044
244
+ Ask for a topic and you
245
+ get its yearly volume.
246
+
247
+ 53
248
+ 00:03:29.084 --> 00:03:33.396
249
+ Large language models: six
250
+ papers in two thousand nineteen,
251
+
252
+ 54
253
+ 00:03:33.436 --> 00:03:37.272
254
+ two hundred and seventy in
255
+ two thousand twenty-five.
256
+
257
+ 55
258
+ 00:03:37.872 --> 00:03:39.424
259
+ But the corpus grew too.
260
+
261
+ 56
262
+ 00:03:39.464 --> 00:03:43.072
263
+ So the second chart divides
264
+ by each year's corpus size.
265
+
266
+ 57
267
+ 00:03:43.112 --> 00:03:48.112
268
+ That is the difference between a topic
269
+ that is genuinely taking over a field
270
+
271
+ 58
272
+ 00:03:48.152 --> 00:03:51.336
273
+ and a topic that is merely
274
+ riding its expansion.
275
+
276
+ 59
277
+ 00:03:51.936 --> 00:03:55.360
278
+ And it names the venues
279
+ carrying that growth,
280
+
281
+ 60
282
+ 00:03:55.400 --> 00:03:58.440
283
+ so the next question, where do I submit,
284
+
285
+ 61
286
+ 00:03:58.480 --> 00:04:03.792
287
+ or where do I read, has an answer
288
+ grounded in the same frozen corpus.
289
+
290
+ 62
291
+ 00:04:04.392 --> 00:04:06.536
292
+ The fourth pass is people.
293
+
294
+ 63
295
+ 00:04:06.576 --> 00:04:13.004
296
+ Researcher Radar ranks who recurs, and
297
+ it is deliberate about what it is not:
298
+
299
+ 64
300
+ 00:04:13.044 --> 00:04:16.028
301
+ it measures presence in this corpus,
302
+
303
+ 65
304
+ 00:04:16.068 --> 00:04:20.184
305
+ never citations, quality,
306
+ seniority or authority.
307
+
308
+ 66
309
+ 00:04:20.784 --> 00:04:25.986
310
+ Three rankings answer three different
311
+ questions. Paper count finds volume.
312
+
313
+ 67
314
+ 00:04:26.026 --> 00:04:29.103
315
+ Tier weighting favours
316
+ the strongest venues.
317
+
318
+ 68
319
+ 00:04:29.143 --> 00:04:32.787
320
+ Top-four concentration asks
321
+ something else entirely:
322
+
323
+ 69
324
+ 00:04:32.827 --> 00:04:38.352
325
+ what share of an author's work lands at
326
+ the four venues that define the field.
327
+
328
+ 70
329
+ 00:04:38.952 --> 00:04:41.235
330
+ Authorship position asks a third.
331
+
332
+ 71
333
+ 00:04:41.275 --> 00:04:44.614
334
+ First author surfaces
335
+ the people doing the work;
336
+
337
+ 72
338
+ 00:04:44.654 --> 00:04:47.781
339
+ last author surfaces
340
+ the people directing it.
341
+
342
+ 73
343
+ 00:04:47.821 --> 00:04:53.664
344
+ For building a reference list, or a program
345
+ committee, those are not the same list.
346
+
347
+ 74
348
+ 00:04:54.264 --> 00:04:56.044
349
+ And no row is a verdict.
350
+
351
+ 75
352
+ 00:04:56.084 --> 00:05:00.593
353
+ Open one and you get the
354
+ trajectory behind it, year by year,
355
+
356
+ 76
357
+ 00:05:00.633 --> 00:05:05.825
358
+ split by byline position, with the
359
+ direct collaborations it rests on.
360
+
361
+ 77
362
+ 00:05:05.865 --> 00:05:08.064
363
+ The heuristic shows its work.
364
+
365
+ 78
366
+ 00:05:08.664 --> 00:05:11.208
367
+ Two derived views close the pass.
368
+
369
+ 79
370
+ 00:05:11.248 --> 00:05:14.809
371
+ Authors whose byline
372
+ moved from first to last,
373
+
374
+ 80
375
+ 00:05:14.849 --> 00:05:18.019
376
+ which is where new groups become visible.
377
+
378
+ 81
379
+ 00:05:18.059 --> 00:05:21.151
380
+ And authors whose annual rate is rising.
381
+
382
+ 82
383
+ 00:05:21.191 --> 00:05:26.280
384
+ Both are described as shortlists
385
+ to investigate, not as findings.
386
+
387
+ 83
388
+ 00:05:26.880 --> 00:05:32.379
389
+ The fourth pass is the one most tools
390
+ omit: where the corpus cannot speak.
391
+
392
+ 84
393
+ 00:05:32.419 --> 00:05:36.721
394
+ Abstract coverage per venue,
395
+ published rather than hidden.
396
+
397
+ 85
398
+ 00:05:36.761 --> 00:05:42.600
399
+ ESORICS sits at six percent, because
400
+ Springer does not expose those abstracts.
401
+
402
+ 86
403
+ 00:05:43.200 --> 00:05:48.846
404
+ That matters, because an abstract search
405
+ cannot include what it cannot read.
406
+
407
+ 87
408
+ 00:05:48.886 --> 00:05:52.063
409
+ A gap you can see is
410
+ an inclusion decision.
411
+
412
+ 88
413
+ 00:05:52.103 --> 00:05:56.592
414
+ A gap you cannot see is a silent
415
+ exclusion in your protocol.
416
+
417
+ 89
418
+ 00:05:57.192 --> 00:06:02.299
419
+ The Evidence page states what this
420
+ snapshot verifies and what it does not.
421
+
422
+ 90
423
+ 00:06:02.339 --> 00:06:04.664
424
+ Abstract quality was not asserted;
425
+
426
+ 91
427
+ 00:06:04.704 --> 00:06:07.793
428
+ it was measured, on a
429
+ venue-stratified sample
430
+
431
+ 92
432
+ 00:06:07.833 --> 00:06:10.824
433
+ of two hundred records
434
+ reviewed by a human.
435
+
436
+ 93
437
+ 00:06:11.424 --> 00:06:14.288
438
+ One hundred and sixty-nine were complete,
439
+
440
+ 94
441
+ 00:06:14.328 --> 00:06:17.545
442
+ uncontaminated, and
443
+ matched the sampled paper.
444
+
445
+ 95
446
+ 00:06:17.585 --> 00:06:19.741
447
+ Eighty-four and a half percent,
448
+
449
+ 96
450
+ 00:06:19.781 --> 00:06:23.069
451
+ with a ninety-five percent
452
+ Wilson interval from
453
+
454
+ 97
455
+ 00:06:23.109 --> 00:06:26.822
456
+ seventy-eight point eight
457
+ to eighty-eight point nine.
458
+
459
+ 98
460
+ 00:06:26.862 --> 00:06:30.120
461
+ The uncertainty is
462
+ reported, not rounded away.
463
+
464
+ 99
465
+ 00:06:30.720 --> 00:06:34.640
466
+ And the release cannot be
467
+ edited from the interface.
468
+
469
+ 100
470
+ 00:06:34.680 --> 00:06:38.066
471
+ Refreshing creates a
472
+ named successor profile;
473
+
474
+ 101
475
+ 00:06:38.106 --> 00:06:42.102
476
+ the snapshot a paper cites
477
+ is never mutated in place.
478
+
479
+ 102
480
+ 00:06:42.142 --> 00:06:45.264
481
+ That is what makes a hash worth printing.
482
+
483
+ 103
484
+ 00:06:45.864 --> 00:06:50.434
485
+ So what does this buy a researcher
486
+ who already knows their field.
487
+
488
+ 104
489
+ 00:06:50.474 --> 00:06:53.808
490
+ Three things, and none
491
+ of them is a search box.
492
+
493
+ 105
494
+ 00:06:54.408 --> 00:06:57.794
495
+ A denominator you can put
496
+ in a paper, with a hash,
497
+
498
+ 106
499
+ 00:06:57.834 --> 00:07:01.426
500
+ that a reviewer can reproduce
501
+ offline in one command.
502
+
503
+ 107
504
+ 00:07:01.466 --> 00:07:05.606
505
+ The methods section stops being
506
+ a description of what you did
507
+
508
+ 108
509
+ 00:07:05.646 --> 00:07:08.592
510
+ and becomes something
511
+ someone else can run.
512
+
513
+ 109
514
+ 00:07:09.192 --> 00:07:11.282
515
+ A fixed corpus turns questions
516
+
517
+ 110
518
+ 00:07:11.322 --> 00:07:16.892
519
+ that used to take a week of manual
520
+ counting into questions you ask in a second.
521
+
522
+ 111
523
+ 00:07:16.932 --> 00:07:19.804
524
+ Topic dynamics normalized against growth.
525
+
526
+ 112
527
+ 00:07:19.844 --> 00:07:22.432
528
+ Author concentration at the top four.
529
+
530
+ 113
531
+ 00:07:22.472 --> 00:07:24.744
532
+ Byline movement across a decade.
533
+
534
+ 114
535
+ 00:07:25.344 --> 00:07:30.851
536
+ And every number on every screen
537
+ opens into the records underneath it.
538
+
539
+ 115
540
+ 00:07:30.891 --> 00:07:33.070
541
+ Nothing asks for your trust.
542
+
543
+ 116
544
+ 00:07:33.110 --> 00:07:37.944
545
+ That is the difference between
546
+ a dashboard and an instrument.
547
+
548
+ 117
549
+ 00:07:38.544 --> 00:07:39.377
550
+ TopVenues.
551
+
552
+ 118
553
+ 00:07:39.417 --> 00:07:44.353
554
+ Open source, MIT licensed,
555
+ with the corpus, the manifest,
556
+
557
+ 119
558
+ 00:07:44.393 --> 00:07:48.408
559
+ and the reproduction
560
+ script in one repository.
assets/demo/topvenues-demo-v1.12.0.jpg ADDED

Git LFS Details

  • SHA256: 2ca1c6e845d3e5d21573dbb4fdfe4af784fe680c8940941a7af095b3ea464137
  • Pointer size: 131 Bytes
  • Size of remote file: 107 kB
assets/demo/topvenues-demo-v1.12.0.mp4 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d8379c54cb44833fccc13b1ccbcabe96c7edfc25a712b97d93f8b38829d59810
3
+ size 32172505
assets/demo/topvenues-demo-v1.12.0.pt-BR.vtt ADDED
@@ -0,0 +1,587 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ WEBVTT
2
+
3
+ 1
4
+ 00:00:00.000 --> 00:00:03.472
5
+ Toda revisão sistemática
6
+ se apoia em um número
7
+
8
+ 2
9
+ 00:00:03.512 --> 00:00:06.831
10
+ que quase nunca é
11
+ registrado: o denominador.
12
+
13
+ 3
14
+ 00:00:06.871 --> 00:00:11.376
15
+ O conjunto de artigos elegíveis
16
+ antes de a triagem começar.
17
+
18
+ 4
19
+ 00:00:11.976 --> 00:00:14.745
20
+ Na prática, esse conjunto
21
+ é remontado à mão,
22
+
23
+ 5
24
+ 00:00:14.785 --> 00:00:17.746
25
+ a partir de portais e
26
+ interfaces cuja cobertura
27
+
28
+ 6
29
+ 00:00:17.786 --> 00:00:20.683
30
+ e semântica de consulta
31
+ mudam sob os seus pés.
32
+
33
+ 7
34
+ 00:00:20.723 --> 00:00:25.320
35
+ Duas pessoas seguem o mesmo protocolo
36
+ e terminam com corpora diferentes.
37
+
38
+ 8
39
+ 00:00:25.920 --> 00:00:28.842
40
+ O TopVenues transforma esse
41
+ denominador em um objeto.
42
+
43
+ 9
44
+ 00:00:28.882 --> 00:00:31.133
45
+ Declarado, congelado, com hash e citável.
46
+
47
+ 10
48
+ 00:00:31.173 --> 00:00:33.240
49
+ É assim que se trabalha com um deles.
50
+
51
+ 11
52
+ 00:00:33.840 --> 00:00:36.505
53
+ A instalação é um clone e um script.
54
+
55
+ 12
56
+ 00:00:36.545 --> 00:00:40.112
57
+ Nada é buscado em editora
58
+ no momento da revisão,
59
+
60
+ 13
61
+ 00:00:40.152 --> 00:00:43.608
62
+ e nada depende de um
63
+ serviço permanecer no ar.
64
+
65
+ 14
66
+ 00:00:44.208 --> 00:00:47.012
67
+ O script cria um ambiente
68
+ isolado a partir de
69
+
70
+ 15
71
+ 00:00:47.052 --> 00:00:49.793
72
+ dependências fixadas
73
+ e verificadas por hash,
74
+
75
+ 16
76
+ 00:00:49.833 --> 00:00:53.774
77
+ e então materializa o corpus a
78
+ partir de um snapshot comprimido
79
+
80
+ 17
81
+ 00:00:53.814 --> 00:00:55.584
82
+ que acompanha o repositório.
83
+
84
+ 18
85
+ 00:00:56.184 --> 00:00:58.714
86
+ Em seguida ele verifica,
87
+ em vez de afirmar.
88
+
89
+ 19
90
+ 00:00:58.754 --> 00:01:01.883
91
+ O SHA-256 do snapshot é
92
+ conferido contra o manifesto,
93
+
94
+ 20
95
+ 00:01:01.923 --> 00:01:06.605
96
+ as contagens de registros são conferidas
97
+ contra os números impressos no artigo,
98
+
99
+ 21
100
+ 00:01:06.645 --> 00:01:09.096
101
+ e a suíte de testes completa é executada.
102
+
103
+ 22
104
+ 00:01:09.696 --> 00:01:12.678
105
+ Vinte mil trezentos e cinco registros.
106
+
107
+ 23
108
+ 00:01:12.718 --> 00:01:18.723
109
+ Dezessete mil quatrocentos e noventa
110
+ e um resumos. Todos os testes passando.
111
+
112
+ 24
113
+ 00:01:18.763 --> 00:01:23.376
114
+ Um revisor chega a esse estado
115
+ sem pedir acesso a ninguém.
116
+
117
+ 25
118
+ 00:01:23.976 --> 00:01:26.773
119
+ A interface abre no que você tem em mãos.
120
+
121
+ 26
122
+ 00:01:26.813 --> 00:01:32.240
123
+ Catorze mil oitocentos e cinquenta e
124
+ nove artigos em vinte veículos declarados,
125
+
126
+ 27
127
+ 00:01:32.280 --> 00:01:35.701
128
+ noventa e quatro por cento
129
+ com resumo pesquisável,
130
+
131
+ 28
132
+ 00:01:35.741 --> 00:01:38.232
133
+ e todo registro pronto para citação.
134
+
135
+ 29
136
+ 00:01:38.832 --> 00:01:40.715
137
+ Nada aqui é consulta ao vivo.
138
+
139
+ 30
140
+ 00:01:40.755 --> 00:01:44.231
141
+ A lista de veículos é um
142
+ escopo científico declarado,
143
+
144
+ 31
145
+ 00:01:44.271 --> 00:01:48.648
146
+ e alterá-la é uma decisão de
147
+ versão, não um filtro que se arrasta.
148
+
149
+ 32
150
+ 00:01:49.248 --> 00:01:51.359
151
+ Comece com uma pergunta comum.
152
+
153
+ 33
154
+ 00:01:51.399 --> 00:01:55.088
155
+ A busca ranqueada ordena
156
+ correspondências de título,
157
+
158
+ 34
159
+ 00:01:55.128 --> 00:02:00.792
160
+ resumo e autor por relevância, sobre
161
+ todo o corpus congelado, em milissegundos.
162
+
163
+ 35
164
+ 00:02:01.392 --> 00:02:03.880
165
+ Restrinja como um protocolo restringe:
166
+
167
+ 36
168
+ 00:02:03.920 --> 00:02:07.606
169
+ por estrato de veículo, por
170
+ ano, por posição de autoria.
171
+
172
+ 37
173
+ 00:02:07.646 --> 00:02:10.933
174
+ A contagem se atualiza,
175
+ e essa contagem é o número
176
+
177
+ 38
178
+ 00:02:10.973 --> 00:02:13.368
179
+ que você reportará como denominador.
180
+
181
+ 39
182
+ 00:02:13.968 --> 00:02:17.620
183
+ Cada linha abre no registro por trás dela.
184
+
185
+ 40
186
+ 00:02:17.660 --> 00:02:24.212
187
+ O resumo que correspondeu, o veículo,
188
+ o estrato, o DOI e o link da editora.
189
+
190
+ 41
191
+ 00:02:24.252 --> 00:02:27.768
192
+ Você lê a evidência, não um resumo dela.
193
+
194
+ 42
195
+ 00:02:28.368 --> 00:02:29.099
196
+ E ele sai.
197
+
198
+ 43
199
+ 00:02:29.139 --> 00:02:35.420
200
+ CSV para uma planilha de triagem, JSON
201
+ para um pipeline, BibTeX para o manuscrito,
202
+
203
+ 44
204
+ 00:02:35.460 --> 00:02:41.088
205
+ extraído diretamente das entradas
206
+ canônicas do DBLP em vez de redigitado.
207
+
208
+ 45
209
+ 00:02:41.688 --> 00:02:45.654
210
+ A aba Insights é onde o
211
+ corpus deixa de ser uma lista.
212
+
213
+ 46
214
+ 00:02:45.694 --> 00:02:47.880
215
+ Ela se lê em quatro passagens:
216
+
217
+ 47
218
+ 00:02:47.920 --> 00:02:50.922
219
+ o que ela contém, como um tópico se move,
220
+
221
+ 48
222
+ 00:02:50.962 --> 00:02:55.488
223
+ quem produz esse trabalho, e
224
+ onde ela não consegue responder.
225
+
226
+ 49
227
+ 00:02:56.088 --> 00:02:57.436
228
+ Composição primeiro.
229
+
230
+ 50
231
+ 00:02:57.476 --> 00:03:01.184
232
+ Artigos por veículo, por
233
+ ano e por classe de registro.
234
+
235
+ 51
236
+ 00:03:01.224 --> 00:03:05.139
237
+ Artigos completos são setenta
238
+ e oito por cento do corpus;
239
+
240
+ 52
241
+ 00:03:05.179 --> 00:03:10.176
242
+ as classes raras são as que um
243
+ protocolo precisa decidir explicitamente.
244
+
245
+ 53
246
+ 00:03:10.776 --> 00:03:12.435
247
+ Todo agregado é uma porta.
248
+
249
+ 54
250
+ 00:03:12.475 --> 00:03:17.598
251
+ Clique numa barra e você está na busca,
252
+ filtrado exatamente naqueles registros.
253
+
254
+ 55
255
+ 00:03:17.638 --> 00:03:21.624
256
+ Nenhum gráfico aqui é uma imagem
257
+ em que você precise confiar.
258
+
259
+ 56
260
+ 00:03:22.224 --> 00:03:25.698
261
+ Agora a passagem que
262
+ muda como se lê uma área.
263
+
264
+ 57
265
+ 00:03:25.738 --> 00:03:29.211
266
+ Peça um tópico e você
267
+ recebe seu volume anual.
268
+
269
+ 58
270
+ 00:03:29.251 --> 00:03:33.642
271
+ Modelos de linguagem: seis
272
+ artigos em dois mil e dezenove,
273
+
274
+ 59
275
+ 00:03:33.682 --> 00:03:37.272
276
+ duzentos e setenta em
277
+ dois mil e vinte e cinco.
278
+
279
+ 60
280
+ 00:03:37.872 --> 00:03:39.689
281
+ Mas o corpus também cresceu.
282
+
283
+ 61
284
+ 00:03:39.729 --> 00:03:44.266
285
+ Por isso o segundo gráfico divide
286
+ pelo tamanho do corpus de cada ano.
287
+
288
+ 62
289
+ 00:03:44.306 --> 00:03:48.510
290
+ Essa é a diferença entre um tópico
291
+ que realmente domina uma área
292
+
293
+ 63
294
+ 00:03:48.550 --> 00:03:51.336
295
+ e um que apenas acompanha a expansão dela.
296
+
297
+ 64
298
+ 00:03:51.936 --> 00:03:55.944
299
+ E ele nomeia os veículos que
300
+ sustentam esse crescimento,
301
+
302
+ 65
303
+ 00:03:55.984 --> 00:03:59.125
304
+ para que a próxima
305
+ pergunta — onde submeter,
306
+
307
+ 66
308
+ 00:03:59.165 --> 00:04:03.792
309
+ ou onde ler — tenha resposta
310
+ ancorada no mesmo corpus congelado.
311
+
312
+ 67
313
+ 00:04:04.392 --> 00:04:07.052
314
+ A quarta passagem são as pessoas.
315
+
316
+ 68
317
+ 00:04:07.092 --> 00:04:13.516
318
+ O Researcher Radar ranqueia quem recorre,
319
+ e é deliberado sobre o que ele não é:
320
+
321
+ 69
322
+ 00:04:13.556 --> 00:04:20.184
323
+ mede presença neste corpus, nunca citações,
324
+ qualidade, senioridade ou autoridade.
325
+
326
+ 70
327
+ 00:04:20.784 --> 00:04:24.244
328
+ Três rankings respondem a
329
+ três perguntas diferentes.
330
+
331
+ 71
332
+ 00:04:24.284 --> 00:04:26.667
333
+ Contagem de artigos encontra volume.
334
+
335
+ 72
336
+ 00:04:26.707 --> 00:04:30.437
337
+ Ponderação por estrato favorece
338
+ os veículos mais fortes.
339
+
340
+ 73
341
+ 00:04:30.477 --> 00:04:33.331
342
+ Concentração no top-4
343
+ pergunta outra coisa:
344
+
345
+ 74
346
+ 00:04:33.371 --> 00:04:38.352
347
+ que fração da obra de um autor cai
348
+ nos quatro veículos que definem a área.
349
+
350
+ 75
351
+ 00:04:38.952 --> 00:04:41.841
352
+ A posição de autoria
353
+ pergunta uma terceira.
354
+
355
+ 76
356
+ 00:04:41.881 --> 00:04:47.358
357
+ Primeiro autor revela quem executa o
358
+ trabalho; último autor revela quem o dirige.
359
+
360
+ 77
361
+ 00:04:47.398 --> 00:04:49.878
362
+ Para montar uma lista de referências,
363
+
364
+ 78
365
+ 00:04:49.918 --> 00:04:53.664
366
+ ou um comitê de programa,
367
+ essas listas não são a mesma.
368
+
369
+ 79
370
+ 00:04:54.264 --> 00:04:56.326
371
+ E nenhuma linha é um veredito.
372
+
373
+ 80
374
+ 00:04:56.366 --> 00:05:00.318
375
+ Abra uma e você vê a trajetória
376
+ por trás dela, ano a ano,
377
+
378
+ 81
379
+ 00:05:00.358 --> 00:05:05.922
380
+ separada por posição na assinatura, com
381
+ as colaborações diretas em que se apoia.
382
+
383
+ 82
384
+ 00:05:05.962 --> 00:05:08.064
385
+ A heurística mostra sua conta.
386
+
387
+ 83
388
+ 00:05:08.664 --> 00:05:11.572
389
+ Duas visões derivadas fecham a passagem.
390
+
391
+ 84
392
+ 00:05:11.612 --> 00:05:15.626
393
+ Autores cuja assinatura
394
+ migrou de primeiro para último,
395
+
396
+ 85
397
+ 00:05:15.666 --> 00:05:18.501
398
+ que é onde novos grupos ficam visíveis.
399
+
400
+ 86
401
+ 00:05:18.541 --> 00:05:21.375
402
+ E autores cuja taxa anual está subindo.
403
+
404
+ 87
405
+ 00:05:21.415 --> 00:05:26.280
406
+ Ambas são descritas como listas
407
+ para investigar, não como achados.
408
+
409
+ 88
410
+ 00:05:26.880 --> 00:05:30.753
411
+ A quarta passagem é a que a
412
+ maioria das ferramentas omite:
413
+
414
+ 89
415
+ 00:05:30.793 --> 00:05:32.980
416
+ onde o corpus não consegue falar.
417
+
418
+ 90
419
+ 00:05:33.020 --> 00:05:37.298
420
+ Cobertura de resumos por veículo,
421
+ publicada em vez de escondida.
422
+
423
+ 91
424
+ 00:05:37.338 --> 00:05:42.600
425
+ O ESORICS está em seis por cento, porque
426
+ a Springer não expõe aqueles resumos.
427
+
428
+ 92
429
+ 00:05:43.200 --> 00:05:48.909
430
+ Isso importa, porque uma busca em resumos
431
+ não pode incluir o que não consegue ler.
432
+
433
+ 93
434
+ 00:05:48.949 --> 00:05:52.065
435
+ Uma lacuna visível é
436
+ uma decisão de inclusão.
437
+
438
+ 94
439
+ 00:05:52.105 --> 00:05:56.592
440
+ Uma lacuna invisível é uma exclusão
441
+ silenciosa no seu protocolo.
442
+
443
+ 95
444
+ 00:05:57.192 --> 00:06:01.883
445
+ A página Evidence declara o que este
446
+ snapshot verifica e o que não verifica.
447
+
448
+ 96
449
+ 00:06:01.923 --> 00:06:04.435
450
+ A qualidade dos resumos não foi afirmada;
451
+
452
+ 97
453
+ 00:06:04.475 --> 00:06:07.360
454
+ foi medida, sobre uma
455
+ amostra estratificada por
456
+
457
+ 98
458
+ 00:06:07.400 --> 00:06:10.824
459
+ veículo de duzentos registros
460
+ revisados por uma pessoa.
461
+
462
+ 99
463
+ 00:06:11.424 --> 00:06:14.120
464
+ Cento e sessenta e nove estavam completos,
465
+
466
+ 100
467
+ 00:06:14.160 --> 00:06:17.638
468
+ não contaminados, e
469
+ correspondiam ao artigo amostrado.
470
+
471
+ 101
472
+ 00:06:17.678 --> 00:06:19.853
473
+ Oitenta e quatro e meio por cento,
474
+
475
+ 102
476
+ 00:06:19.893 --> 00:06:24.152
477
+ com intervalo de Wilson de noventa
478
+ e cinco por cento entre setenta
479
+
480
+ 103
481
+ 00:06:24.192 --> 00:06:27.409
482
+ e oito vírgula oito e
483
+ oitenta e oito vírgula nove.
484
+
485
+ 104
486
+ 00:06:27.449 --> 00:06:30.120
487
+ A incerteza é reportada, não arredondada.
488
+
489
+ 105
490
+ 00:06:30.720 --> 00:06:34.132
491
+ E a versão não pode ser
492
+ editada pela interface.
493
+
494
+ 106
495
+ 00:06:34.172 --> 00:06:37.658
496
+ Uma atualização cria um
497
+ perfil sucessor nomeado;
498
+
499
+ 107
500
+ 00:06:37.698 --> 00:06:41.772
501
+ o snapshot que um artigo cita
502
+ nunca é alterado no lugar.
503
+
504
+ 108
505
+ 00:06:41.812 --> 00:06:45.264
506
+ É isso que torna um hash
507
+ digno de ser impresso.
508
+
509
+ 109
510
+ 00:06:45.864 --> 00:06:50.344
511
+ Então o que isso oferece a um
512
+ pesquisador que já conhece sua área?
513
+
514
+ 110
515
+ 00:06:50.384 --> 00:06:53.808
516
+ Três coisas, e nenhuma
517
+ delas é uma caixa de busca.
518
+
519
+ 111
520
+ 00:06:54.408 --> 00:06:57.564
521
+ Um denominador que você
522
+ pode colocar num artigo,
523
+
524
+ 112
525
+ 00:06:57.604 --> 00:07:01.294
526
+ com hash, que um revisor
527
+ reproduz offline em um comando.
528
+
529
+ 113
530
+ 00:07:01.334 --> 00:07:05.222
531
+ A seção de métodos deixa de
532
+ ser a descrição do que você fez
533
+
534
+ 114
535
+ 00:07:05.262 --> 00:07:08.592
536
+ e passa a ser algo que
537
+ outra pessoa pode executar.
538
+
539
+ 115
540
+ 00:07:09.192 --> 00:07:11.374
541
+ Um corpus fixo transforma perguntas
542
+
543
+ 116
544
+ 00:07:11.414 --> 00:07:16.515
545
+ que levavam uma semana de contagem manual
546
+ em perguntas respondidas em um segundo.
547
+
548
+ 117
549
+ 00:07:16.555 --> 00:07:19.626
550
+ Dinâmica de tópicos
551
+ normalizada pelo crescimento.
552
+
553
+ 118
554
+ 00:07:19.666 --> 00:07:21.721
555
+ Concentração de autores no top-4.
556
+
557
+ 119
558
+ 00:07:21.761 --> 00:07:24.744
559
+ Movimento de assinatura
560
+ ao longo de uma década.
561
+
562
+ 120
563
+ 00:07:25.344 --> 00:07:31.079
564
+ E todo número em toda tela abre
565
+ nos registros que estão por baixo.
566
+
567
+ 121
568
+ 00:07:31.119 --> 00:07:33.354
569
+ Nada pede a sua confiança.
570
+
571
+ 122
572
+ 00:07:33.394 --> 00:07:37.944
573
+ Essa é a diferença entre
574
+ um painel e um instrumento.
575
+
576
+ 123
577
+ 00:07:38.544 --> 00:07:39.385
578
+ TopVenues.
579
+
580
+ 124
581
+ 00:07:39.425 --> 00:07:42.996
582
+ Código aberto, licença MIT, com o corpus,
583
+
584
+ 125
585
+ 00:07:43.036 --> 00:07:48.408
586
+ o manifesto e o script de
587
+ reprodução em um único repositório.