-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathindex.html
1244 lines (853 loc) · 63.1 KB
/
index.html
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
<!DOCTYPE html>
<html lang="zh-CN">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1, maximum-scale=2">
<meta name="theme-color" content="#222">
<meta name="generator" content="Hexo 6.0.0">
<link rel="apple-touch-icon" sizes="180x180" href="/images/apple-touch-icon-next.png">
<link rel="icon" type="image/png" sizes="32x32" href="/images/favicon-32x32-next.png">
<link rel="icon" type="image/png" sizes="16x16" href="/images/favicon-16x16-next.png">
<link rel="mask-icon" href="/images/logo.svg" color="#222">
<link rel="stylesheet" href="/css/main.css">
<link rel="stylesheet" href="/lib/font-awesome/css/all.min.css">
<script id="hexo-configurations">
var NexT = window.NexT || {};
var CONFIG = {"hostname":"dawnzzz.github.io","root":"/","scheme":"Gemini","version":"7.8.0","exturl":false,"sidebar":{"position":"left","display":"post","padding":18,"offset":12,"onmobile":false},"copycode":{"enable":false,"show_result":false,"style":null},"back2top":{"enable":true,"sidebar":false,"scrollpercent":false},"bookmark":{"enable":false,"color":"#222","save":"auto"},"fancybox":false,"mediumzoom":false,"lazyload":false,"pangu":false,"comments":{"style":"tabs","active":null,"storage":true,"lazyload":false,"nav":null},"algolia":{"hits":{"per_page":10},"labels":{"input_placeholder":"Search for Posts","hits_empty":"We didn't find any results for the search: ${query}","hits_stats":"${hits} results found in ${time} ms"}},"localsearch":{"enable":false,"trigger":"auto","top_n_per_article":1,"unescape":false,"preload":false},"motion":{"enable":true,"async":false,"transition":{"post_block":"fadeIn","post_header":"slideDownIn","post_body":"slideDownIn","coll_header":"slideLeftIn","sidebar":"slideUpIn"}}};
</script>
<meta name="description" content="个人博客站点">
<meta property="og:type" content="website">
<meta property="og:title" content="Dawn's Blogs">
<meta property="og:url" content="http://dawnzzz.github.io/index.html">
<meta property="og:site_name" content="Dawn's Blogs">
<meta property="og:description" content="个人博客站点">
<meta property="og:locale" content="zh_CN">
<meta property="article:author" content="DawnZH">
<meta name="twitter:card" content="summary">
<link rel="canonical" href="http://dawnzzz.github.io/">
<script id="page-configurations">
// https://hexo.io/docs/variables.html
CONFIG.page = {
sidebar: "",
isHome : true,
isPost : false,
lang : 'zh-CN'
};
</script>
<title>Dawn's Blogs</title>
<noscript>
<style>
.use-motion .brand,
.use-motion .menu-item,
.sidebar-inner,
.use-motion .post-block,
.use-motion .pagination,
.use-motion .comments,
.use-motion .post-header,
.use-motion .post-body,
.use-motion .collection-header { opacity: initial; }
.use-motion .site-title,
.use-motion .site-subtitle {
opacity: initial;
top: initial;
}
.use-motion .logo-line-before i { left: initial; }
.use-motion .logo-line-after i { right: initial; }
</style>
</noscript>
</head>
<body itemscope itemtype="http://schema.org/WebPage">
<div class="container use-motion">
<div class="headband"></div>
<header class="header" itemscope itemtype="http://schema.org/WPHeader">
<div class="header-inner"><div class="site-brand-container">
<div class="site-nav-toggle">
<div class="toggle" aria-label="切换导航栏">
<span class="toggle-line toggle-line-first"></span>
<span class="toggle-line toggle-line-middle"></span>
<span class="toggle-line toggle-line-last"></span>
</div>
</div>
<div class="site-meta">
<a href="/" class="brand" rel="start">
<span class="logo-line-before"><i></i></span>
<h1 class="site-title">Dawn's Blogs</h1>
<span class="logo-line-after"><i></i></span>
</a>
<p class="site-subtitle" itemprop="description">分享技术 记录成长</p>
</div>
<div class="site-nav-right">
<div class="toggle popup-trigger">
</div>
</div>
</div>
<nav class="site-nav">
<ul id="menu" class="main-menu menu">
<li class="menu-item menu-item-home">
<a href="/" rel="section"><i class="fa fa-home fa-fw"></i>首页</a>
</li>
<li class="menu-item menu-item-categories">
<a href="/categories/" rel="section"><i class="fa fa-th fa-fw"></i>分类</a>
</li>
<li class="menu-item menu-item-tags">
<a href="/tags/" rel="section"><i class="fa fa-tags fa-fw"></i>标签</a>
</li>
<li class="menu-item menu-item-archives">
<a href="/archives/" rel="section"><i class="fa fa-archive fa-fw"></i>归档</a>
</li>
</ul>
</nav>
</div>
</header>
<div class="back-to-top">
<i class="fa fa-arrow-up"></i>
<span>0%</span>
</div>
<main class="main">
<div class="main-inner">
<div class="content-wrap">
<div class="content index posts-expand">
<article itemscope itemtype="http://schema.org/Article" class="post-block" lang="zh-CN">
<link itemprop="mainEntityOfPage" href="http://dawnzzz.github.io/2024/10/02/%E5%85%B3%E7%B3%BB%E6%8A%BD%E5%8F%96%E8%AE%BA%E6%96%87-Document-Level-Relation-Extraction-with-Adaptive-Focal-Loss-and-Knowledge-Distillation/">
<span hidden itemprop="author" itemscope itemtype="http://schema.org/Person">
<meta itemprop="image" content="/images/avatar.gif">
<meta itemprop="name" content="DawnZH">
<meta itemprop="description" content="个人博客站点">
</span>
<span hidden itemprop="publisher" itemscope itemtype="http://schema.org/Organization">
<meta itemprop="name" content="Dawn's Blogs">
</span>
<header class="post-header">
<h2 class="post-title" itemprop="name headline">
<a href="/2024/10/02/%E5%85%B3%E7%B3%BB%E6%8A%BD%E5%8F%96%E8%AE%BA%E6%96%87-Document-Level-Relation-Extraction-with-Adaptive-Focal-Loss-and-Knowledge-Distillation/" class="post-title-link" itemprop="url">关系抽取论文 Document-Level Relation Extraction with Adaptive Focal Loss and Knowledge Distillation</a>
</h2>
<div class="post-meta">
<span class="post-meta-item">
<span class="post-meta-item-icon">
<i class="far fa-calendar"></i>
</span>
<span class="post-meta-item-text">发表于</span>
<time title="创建时间：2024-10-02 17:56:23 / 修改时间：20:08:22" itemprop="dateCreated datePublished" datetime="2024-10-02T17:56:23+08:00">2024-10-02</time>
</span>
<span class="post-meta-item">
<span class="post-meta-item-icon">
<i class="far fa-folder"></i>
</span>
<span class="post-meta-item-text">分类于</span>
<span itemprop="about" itemscope itemtype="http://schema.org/Thing">
<a href="/categories/%E5%85%B3%E7%B3%BB%E6%8A%BD%E5%8F%96%E8%AE%BA%E6%96%87/" itemprop="url" rel="index"><span itemprop="name">关系抽取论文</span></a>
</span>
</span>
</div>
</header>
<div class="post-body" itemprop="articleBody">
<blockquote>
<p>Document-Level Relation Extraction with Adaptive Focal Loss and Knowledge Distillation</p>
<p>年份：2022</p>
<p>From：ACL</p>
<p>作者：Qingyu Tan, Ruidan He etc.</p>
<p>机构：DAMO Academy, Alibaba Group</p>
<p>GitHub：<a target="_blank" rel="noopener" href="https://github.com/tonytan48/KD-DocRE">https://github.com/tonytan48/KD-DocRE</a></p>
<p>工作：</p>
<ul>
<li>为了改进了双跳关系的推理，提出使用<strong>轴向注意力</strong>模块作为特征提取器。这个模块能够关注位于双跳逻辑路径中的元素，并捕获关系三元组之间的相互依赖关系。</li>
<li>其次，提出了自适应焦点损失（Adaptive Focal Loss）来解决不平衡的关系类型分布问题，提出的损失函数使得长尾类对总体损失的贡献更大。</li>
<li>最后，使用知识蒸馏来克服注释数据和远程监督数据之间的差异。具体来说，首先用少量的人类注释数据来训练一个教师模型。然后，将使用教师模型对大量的远程监督数据进行预测。生成的预测被用作预训练学生模型的软标签。最后，对预先训练好的学生模型进行了进一步的微调。</li>
</ul>
</blockquote>
<p>现有工作的缺陷：</p>
<ol>
<li>现有方法聚焦于从预训练语言模型获取上下文信息，但是忽略了实体对之间的交互信息。</li>
<li>现有方法没有明确的解决DocRE数据集中关系类别数量不平衡的问题，仅关注于动态阈值以平衡正负样本数量不平衡的问题。</li>
<li>几乎没有工作讨论在DocRE数据集上应用远程监督方法。</li>
</ol>
<h1 id="模型结构"><a href="#模型结构" class="headerlink" title="模型结构"></a>模型结构</h1><p><img src="/../images/%E5%85%B3%E7%B3%BB%E6%8A%BD%E5%8F%96%E8%AE%BA%E6%96%87-Document-Level-Relation-Extraction-with-Adaptive-Focal-Loss-and-Knowledge-Distillation/image-20241002183958840.png" alt="image-20241002183958840"></p>
<p>提出的模型包含三个部分：</p>
<ol>
<li>表示学习（Representation Learning）：首先通过预先训练的语言模型提取每个实体对的上下文表示。轴向注意模块将进一步增强实体对的表示，该模块将对实体对之间的相互依赖信息进行编码。</li>
<li>自适应Focal损失（Adaptive Focal Loss）：使用一个前馈神经网络分类器，用于计算损失，应用AFL损失以更好的应对长尾分布。</li>
<li>知识蒸馏：使用知识蒸馏客服人工标注数据和远程数据之间的差异。具体来说，就是用带注释的数据训练一个教师模型，并将其输出作为软标签。然后，我们基于软标签和远程标签对学生模型进行预训练。预先训练好的学生模型将再次使用带注释的数据进行微调。</li>
</ol>
<h2 id="Representation-Learning"><a href="#Representation-Learning" class="headerlink" title="Representation Learning"></a>Representation Learning</h2><h3 id="实体表示"><a href="#实体表示" class="headerlink" title="实体表示"></a>实体表示</h3><p>利用预训练语言模型对文档进行编码，同时用特殊token标记实体位置（星号）。实体表示为，多个实体提及的sumexp池化操作：</p>
<p><img src="/../images/%E5%85%B3%E7%B3%BB%E6%8A%BD%E5%8F%96%E8%AE%BA%E6%96%87-Document-Level-Relation-Extraction-with-Adaptive-Focal-Loss-and-Knowledge-Distillation/image-20241002192707135.png" alt="image-20241002192707135"></p>
<p>后用多个实体提及在最后一层的多头注意力权重的平均值，作为上下文信息增强：</p>
<p><img src="/../images/%E5%85%B3%E7%B3%BB%E6%8A%BD%E5%8F%96%E8%AE%BA%E6%96%87-Document-Level-Relation-Extraction-with-Adaptive-Focal-Loss-and-Knowledge-Distillation/image-20241002192807553.png" alt="image-20241002192807553"></p>
<p>最终，实体表示为两部分，分别是实体嵌入和上下文多头注意力机制增强：</p>
<p><img src="/../images/%E5%85%B3%E7%B3%BB%E6%8A%BD%E5%8F%96%E8%AE%BA%E6%96%87-Document-Level-Relation-Extraction-with-Adaptive-Focal-Loss-and-Knowledge-Distillation/image-20241002192855565.png" alt="image-20241002192855565"></p>
<h3 id="实体对表示（轴向注意力）"><a href="#实体对表示（轴向注意力）" class="headerlink" title="实体对表示（轴向注意力）"></a>实体对表示（轴向注意力）</h3><p>实体对通过分组的线性函数进行表示：</p>
<p><img src="/../images/%E5%85%B3%E7%B3%BB%E6%8A%BD%E5%8F%96%E8%AE%BA%E6%96%87-Document-Level-Relation-Extraction-with-Adaptive-Focal-Loss-and-Knowledge-Distillation/image-20241002193001011.png" alt="image-20241002193001011"></p>
<p>引入轴向注意力机制，来引入两跳的实体对信息，最终实体对表示为：</p>
<p><img src="/../images/%E5%85%B3%E7%B3%BB%E6%8A%BD%E5%8F%96%E8%AE%BA%E6%96%87-Document-Level-Relation-Extraction-with-Adaptive-Focal-Loss-and-Knowledge-Distillation/image-20241002193326491.png" alt="image-20241002193326491"></p>
<h2 id="Adaptive-Focus-Loss"><a href="#Adaptive-Focus-Loss" class="headerlink" title="Adaptive Focus Loss"></a>Adaptive Focus Loss</h2><p>通过线性层来进行关系类型的预测：</p>
<p><img src="/../images/%E5%85%B3%E7%B3%BB%E6%8A%BD%E5%8F%96%E8%AE%BA%E6%96%87-Document-Level-Relation-Extraction-with-Adaptive-Focal-Loss-and-Knowledge-Distillation/image-20241002193633475.png" alt="image-20241002193633475"></p>
<p>引入AFL来解决长尾分布问题，损失函数分为两个部分。第一部分是正例的，第二部分是关于负例的。其中，正例的概率分布计算方式：</p>
<p><img src="/../images/%E5%85%B3%E7%B3%BB%E6%8A%BD%E5%8F%96%E8%AE%BA%E6%96%87-Document-Level-Relation-Extraction-with-Adaptive-Focal-Loss-and-Knowledge-Distillation/image-20241002193834274.png" alt="image-20241002193834274"></p>
<p>负例的概率分布计算方式：</p>
<p><img src="/../images/%E5%85%B3%E7%B3%BB%E6%8A%BD%E5%8F%96%E8%AE%BA%E6%96%87-Document-Level-Relation-Extraction-with-Adaptive-Focal-Loss-and-Knowledge-Distillation/image-20241002193856005.png" alt="image-20241002193856005"></p>
<p>损失函数被定义为：</p>
<p><img src="/../images/%E5%85%B3%E7%B3%BB%E6%8A%BD%E5%8F%96%E8%AE%BA%E6%96%87-Document-Level-Relation-Extraction-with-Adaptive-Focal-Loss-and-Knowledge-Distillation/image-20241002194006213.png" alt="image-20241002194006213"></p>
<p>其中r为超参数，AFL损失是被设计成更多地关注低置信度的类别。如果一个正例关系类别的概率分布较低，则相关类的损失贡献将较高，从而可以对长尾类进行更好的优化。</p>
<h2 id="利用远程监督进行知识蒸馏"><a href="#利用远程监督进行知识蒸馏" class="headerlink" title="利用远程监督进行知识蒸馏"></a>利用远程监督进行知识蒸馏</h2><p>略，这部分没有参考价值。</p>
<h1 id="实验"><a href="#实验" class="headerlink" title="实验"></a>实验</h1><p><img src="/../images/%E5%85%B3%E7%B3%BB%E6%8A%BD%E5%8F%96%E8%AE%BA%E6%96%87-Document-Level-Relation-Extraction-with-Adaptive-Focal-Loss-and-Knowledge-Distillation/image-20241002192356297.png" alt="image-20241002192356297"></p>
</div>
<footer class="post-footer">
<div class="post-eof"></div>
</footer>
</article>
<article itemscope itemtype="http://schema.org/Article" class="post-block" lang="zh-CN">
<link itemprop="mainEntityOfPage" href="http://dawnzzz.github.io/2024/10/02/%E5%85%B3%E7%B3%BB%E6%8A%BD%E5%8F%96%E8%AE%BA%E6%96%87-RAPL-A-Relation-Aware-Prototype-Learning-Approach-for-Few-Shot-Document-Level-Relation-Extraction/">
<span hidden itemprop="author" itemscope itemtype="http://schema.org/Person">
<meta itemprop="image" content="/images/avatar.gif">
<meta itemprop="name" content="DawnZH">
<meta itemprop="description" content="个人博客站点">
</span>
<span hidden itemprop="publisher" itemscope itemtype="http://schema.org/Organization">
<meta itemprop="name" content="Dawn's Blogs">
</span>
<header class="post-header">
<h2 class="post-title" itemprop="name headline">
<a href="/2024/10/02/%E5%85%B3%E7%B3%BB%E6%8A%BD%E5%8F%96%E8%AE%BA%E6%96%87-RAPL-A-Relation-Aware-Prototype-Learning-Approach-for-Few-Shot-Document-Level-Relation-Extraction/" class="post-title-link" itemprop="url">关系抽取论文 RAPL: A Relation-Aware Prototype Learning Approach for Few-Shot Document-Level Relation Extraction</a>
</h2>
<div class="post-meta">
<span class="post-meta-item">
<span class="post-meta-item-icon">
<i class="far fa-calendar"></i>
</span>
<span class="post-meta-item-text">发表于</span>
<time title="创建时间：2024-10-02 14:14:03 / 修改时间：16:44:15" itemprop="dateCreated datePublished" datetime="2024-10-02T14:14:03+08:00">2024-10-02</time>
</span>
<span class="post-meta-item">
<span class="post-meta-item-icon">
<i class="far fa-folder"></i>
</span>
<span class="post-meta-item-text">分类于</span>
<span itemprop="about" itemscope itemtype="http://schema.org/Thing">
<a href="/categories/%E5%85%B3%E7%B3%BB%E6%8A%BD%E5%8F%96%E8%AE%BA%E6%96%87/" itemprop="url" rel="index"><span itemprop="name">关系抽取论文</span></a>
</span>
</span>
</div>
</header>
<div class="post-body" itemprop="articleBody">
<blockquote>
<p>RAPL: A Relation-Aware Prototype Learning Approach for Few-Shot Document-Level Relation Extraction</p>
<p>年份：2023</p>
<p>From：EMNLP</p>
<p>作者：Shiao Meng, Xuming Hu, Aiwei Liu, Shu’ang Li, Fukun Ma, Yawen Yang, Lijie Wen</p>
<p>机构：School of Software, Tsinghua University</p>
<p>GitHub：<a target="_blank" rel="noopener" href="https://github.com/THU-BPM/RAPL">https://github.com/THU-BPM/RAPL</a></p>
<p>文档型关系抽取数据的标记费时间、人工密集，同时在垂直领域中缺乏标记数据，文档型关系抽取标记稀缺是现实世界的常见问题。文章提出了RAPL（Relation-Awared Prototype Learning），解决了以下问题：</p>
<ul>
<li><p>对于支持文档中包含关系的每个实体对，<strong>利用关系描述中固有的关系语义作为指导，为每个表达的关系推导出一个实例级表示</strong>。关系原型是通过聚合其所有关系实例的表示来构建的，从而更好地关注与关系相关的信息。基于实例级的嵌入，又提出了一种关系加权对比学习方法来进一步细化原型，可以更好地区分语义密切关系的原型。</p>
</li>
<li><p>设计了一个<strong>特定于任务的NOTA原型生成策略</strong>。对于每个任务，自适应地选择支持的NOTA实例，并将它们融合成一组可学习的基础NOTA原型，以生成特定于任务的NOTA原型，从而更有效地捕获每个任务中的NOTA语义。</p>
</li>
</ul>
<p><img src="/../images/%E5%85%B3%E7%B3%BB%E6%8A%BD%E5%8F%96%E8%AE%BA%E6%96%87-RAPL-A-Relation-Aware-Prototype-Learning-Approach-for-Few-Shot-Document-Level-Relation-Extraction/image-20241002144134220.png" alt="image-20241002144134220"></p>
</blockquote>
<p>few-shot文档级关系抽取在实际应用中，针对labaled文档稀缺问题至关重要。现有的工作主要是matric-based元学习框架，为关系类型建模原型，是FSDLRE（Few-Shot Document-Level RE）中被应用广泛的一种框架。然而，现有的工作往往难以获得具有精确关系语义的类原型，<strong>缺陷</strong>：</p>
<ol>
<li>为了构建目标关系类型的原型，它们聚合了带有该关系的所有实体对的表示，而这些<strong>实体对也可能包含其他关系</strong>，从而干扰了原型（Overlap问题）。</li>
<li>在所有任务中使用<strong>一组通用的NOTA</strong>（None of the above）原型，而忽略了在具有不同目标关系类型的任务中，NOTA语义的不同。</li>
</ol>
<p><img src="/../images/%E5%85%B3%E7%B3%BB%E6%8A%BD%E5%8F%96%E8%AE%BA%E6%96%87-RAPL-A-Relation-Aware-Prototype-Learning-Approach-for-Few-Shot-Document-Level-Relation-Extraction/image-20241002144148114.png" alt="image-20241002144148114"></p>
<h1 id="模型结构"><a href="#模型结构" class="headerlink" title="模型结构"></a>模型结构</h1><p><img src="/../images/%E5%85%B3%E7%B3%BB%E6%8A%BD%E5%8F%96%E8%AE%BA%E6%96%87-RAPL-A-Relation-Aware-Prototype-Learning-Approach-for-Few-Shot-Document-Level-Relation-Extraction/image-20241002144624283.png" alt="image-20241002144624283"></p>
<h2 id="文档和实体编码器"><a href="#文档和实体编码器" class="headerlink" title="文档和实体编码器"></a>文档和实体编码器</h2><h3 id="文档编码"><a href="#文档编码" class="headerlink" title="文档编码"></a>文档编码</h3><p>使用<strong>预训练语言模型</strong>（BERT等）作为文档编码器，对于每一个文档，在每个实体提及的开始和结束处插入一个特殊的token“∗”，以标记实体的位置。然后，将文档输入预训练语言模型编码器，以获得具有上下文信息的token嵌入H和交叉token注意力矩阵A。其中，H是BERT的输出；A是最后一层的平均注意力头。</p>
<p><img src="/../images/%E5%85%B3%E7%B3%BB%E6%8A%BD%E5%8F%96%E8%AE%BA%E6%96%87-RAPL-A-Relation-Aware-Prototype-Learning-Approach-for-Few-Shot-Document-Level-Relation-Extraction/image-20241002145607492.png" alt="image-20241002145607492"></p>
<h3 id="实体编码"><a href="#实体编码" class="headerlink" title="实体编码"></a>实体编码</h3><p>对于一个实体而言，在一个文档中可能有多次提及，使用插入特殊的标记“*”作为提及的嵌入，<strong>实体的嵌入就是多个提及嵌入的sumexp池化</strong>。</p>
<h2 id="关系感知的关系原型学习"><a href="#关系感知的关系原型学习" class="headerlink" title="关系感知的关系原型学习"></a>关系感知的关系原型学习</h2><p>对于给定事件中的每个目标关系类型，目标是获得一个能够更好地捕获相应关系语义的原型表示。为此：</p>
<ul>
<li>首先提出构建基于实例级（一个关系三元组就是一个实例）嵌入的关系原型，使每个原型能够更多地关注支持文档中的关系相关信息。</li>
<li>然后提出了一种实例级的关系加权对比学习方法，进一步细化了关系原型。</li>
</ul>
<h3 id="基于实例的原型构建"><a href="#基于实例的原型构建" class="headerlink" title="基于实例的原型构建"></a>基于实例的原型构建</h3><h4 id="实体对级别的注意力分布"><a href="#实体对级别的注意力分布" class="headerlink" title="实体对级别的注意力分布"></a>实体对级别的注意力分布</h4><p>首先得到<strong>实体pair级别的注意力分布</strong>，获取与实体pair相关的上下文。其中实体级别的注意力分布是实体提及嵌入（A矩阵得到）的平均值：</p>
<p><img src="/../images/%E5%85%B3%E7%B3%BB%E6%8A%BD%E5%8F%96%E8%AE%BA%E6%96%87-RAPL-A-Relation-Aware-Prototype-Learning-Approach-for-Few-Shot-Document-Level-Relation-Extraction/image-20241002151547711.png" alt="image-20241002151547711"></p>
<h4 id="关系级别的注意力分布"><a href="#关系级别的注意力分布" class="headerlink" title="关系级别的注意力分布"></a>关系级别的注意力分布</h4><p>同时，获取<strong>关系级别的注意力分布</strong>，使用另一个预先训练好的语言模型作为关系编码器，并将关系的名称和描述连接到一个序列中，然后将该序列输入编码器。以[CLS]的输出作为关系嵌入，并计算关系级别的注意力分布：</p>
<p><img src="/../images/%E5%85%B3%E7%B3%BB%E6%8A%BD%E5%8F%96%E8%AE%BA%E6%96%87-RAPL-A-Relation-Aware-Prototype-Learning-Approach-for-Few-Shot-Document-Level-Relation-Extraction/image-20241002151945127.png" alt="image-20241002151945127"></p>
<h4 id="实例级别的嵌入计算"><a href="#实例级别的嵌入计算" class="headerlink" title="实例级别的嵌入计算"></a>实例级别的嵌入计算</h4><p>基于实体对级别和关系级别的注意力嵌入，计算<strong>实例级别的注意力嵌入</strong>，以获取实例级别的上下文信息。其中top-k%函数表示获取%k大的下标。利用关系级的注意力来放大与实例最相关的上下文的实体对级别嵌入的权重。</p>
<p><img src="/../images/%E5%85%B3%E7%B3%BB%E6%8A%BD%E5%8F%96%E8%AE%BA%E6%96%87-RAPL-A-Relation-Aware-Prototype-Learning-Approach-for-Few-Shot-Document-Level-Relation-Extraction/image-20241002152119975.png" alt="image-20241002152119975"></p>
<p>最后，计算实例级别的上下文嵌入：</p>
<p><img src="/../images/%E5%85%B3%E7%B3%BB%E6%8A%BD%E5%8F%96%E8%AE%BA%E6%96%87-RAPL-A-Relation-Aware-Prototype-Learning-Approach-for-Few-Shot-Document-Level-Relation-Extraction/image-20241002152533823.png" alt="image-20241002152533823"></p>
<p>并将其融合到头实体和尾实体的嵌入中，获得具有实例感知能力的实体表示：</p>
<p><img src="/../images/%E5%85%B3%E7%B3%BB%E6%8A%BD%E5%8F%96%E8%AE%BA%E6%96%87-RAPL-A-Relation-Aware-Prototype-Learning-Approach-for-Few-Shot-Document-Level-Relation-Extraction/image-20241002152616492.png" alt="image-20241002152616492"></p>
<h4 id="原型计算"><a href="#原型计算" class="headerlink" title="原型计算"></a>原型计算</h4><p>将头实体和尾实体视为实例级别的嵌入：</p>
<p><img src="/../images/%E5%85%B3%E7%B3%BB%E6%8A%BD%E5%8F%96%E8%AE%BA%E6%96%87-RAPL-A-Relation-Aware-Prototype-Learning-Approach-for-Few-Shot-Document-Level-Relation-Extraction/image-20241002152831048.png" alt="image-20241002152831048"></p>
<p>原型为实例级别嵌入的平均值：</p>
<p><img src="/../images/%E5%85%B3%E7%B3%BB%E6%8A%BD%E5%8F%96%E8%AE%BA%E6%96%87-RAPL-A-Relation-Aware-Prototype-Learning-Approach-for-Few-Shot-Document-Level-Relation-Extraction/image-20241002152900562.png" alt="image-20241002152900562"></p>
<h3 id="关系加权对比学习方法"><a href="#关系加权对比学习方法" class="headerlink" title="关系加权对比学习方法"></a>关系加权对比学习方法</h3><p>提出了一种关系加权的对比学习方法以细化关系原型表示，对比学习的目标是：使得同一关系类型的实例级别表示更近，不同关系类型的实力级别表示更远。</p>
<p>S为Support集合中的所有实例集合；对于一个特定的实例（关系三元组），P为表达同一关系的其他实例的集合，A为其他所有实例的集合。然后定义关系加权对比学习的损失函数：</p>
<ul>
<li>分子越大，分数越大，损失函数越小：当前实例和表达同一关系的其他实例的点积越大，故当前实例和表达同一关系的其他实例越接近。</li>
<li>分母越小，分数越大，损失函数越小：引入关系相似度（cossim），当不同关系的语义相似度越接近时具备更大损失，故当前实例和表达不同关系的其他实例越远。</li>
</ul>
<p><img src="/../images/%E5%85%B3%E7%B3%BB%E6%8A%BD%E5%8F%96%E8%AE%BA%E6%96%87-RAPL-A-Relation-Aware-Prototype-Learning-Approach-for-Few-Shot-Document-Level-Relation-Extraction/image-20241002154416464.png" alt="image-20241002154416464"></p>
<blockquote>
<p>通过引入相似性，损失函数可以更好的区分相似语义的不同关系。</p>
</blockquote>
<h2 id="关系感知的NOTA原型学习"><a href="#关系感知的NOTA原型学习" class="headerlink" title="关系感知的NOTA原型学习"></a>关系感知的NOTA原型学习</h2><p>现有方法中，通常学习一组NOTA原型以应用于所有的任务。这在某些任务中可能不是最优的，<strong>因为NOTA语义在具有不同目标关系类型的任务中是不同的</strong>。为此，提出了一种<strong>特定于任务的NOTA原型生成策略</strong>，以更好地捕获每个单独任务中的NOTA语义。</p>
<p>引入一组可学习的向量，与以前直接将这组向量视为NOTA原型的工作不同，将它们视为需要在每个任务中进行进一步修正的基础NOTA原型。由于Support文档的注释是完整的，所以可以获取到Support集合中的隐式表达语义的NOTA分布。因此，依靠Support集合中的NOTA实例来捕获每个特定任务中的NOTA语义。</p>
<p>下面表达式中，它定位了接近基础NOTA原型、并且远离关系原型的NOTA实例：</p>
<p><img src="/../images/%E5%85%B3%E7%B3%BB%E6%8A%BD%E5%8F%96%E8%AE%BA%E6%96%87-RAPL-A-Relation-Aware-Prototype-Learning-Approach-for-Few-Shot-Document-Level-Relation-Extraction/image-20241002161330621.png" alt="image-20241002161330621"></p>
<p>NOTA原型如下，将获取到的NOTA实例融合到基础NOTA原型中：</p>
<p><img src="/../images/%E5%85%B3%E7%B3%BB%E6%8A%BD%E5%8F%96%E8%AE%BA%E6%96%87-RAPL-A-Relation-Aware-Prototype-Learning-Approach-for-Few-Shot-Document-Level-Relation-Extraction/image-20241002161340351.png" alt="image-20241002161340351"></p>
<h2 id="推理-训练过程"><a href="#推理-训练过程" class="headerlink" title="推理/训练过程"></a>推理/训练过程</h2><p>推理时，实体之间的关系类型概率判别如下：</p>
<p><img src="/../images/%E5%85%B3%E7%B3%BB%E6%8A%BD%E5%8F%96%E8%AE%BA%E6%96%87-RAPL-A-Relation-Aware-Prototype-Learning-Approach-for-Few-Shot-Document-Level-Relation-Extraction/image-20241002162118440.png" alt="image-20241002162118440"></p>
<p>在训练时，损失函数不仅仅包含对比学习损失，还包括Query预测结果损失（BCE）：</p>
<p><img src="/../images/%E5%85%B3%E7%B3%BB%E6%8A%BD%E5%8F%96%E8%AE%BA%E6%96%87-RAPL-A-Relation-Aware-Prototype-Learning-Approach-for-Few-Shot-Document-Level-Relation-Extraction/image-20241002162148511.png" alt="image-20241002162148511"></p>
<h1 id="结果"><a href="#结果" class="headerlink" title="结果"></a>结果</h1><p><img src="/../images/%E5%85%B3%E7%B3%BB%E6%8A%BD%E5%8F%96%E8%AE%BA%E6%96%87-RAPL-A-Relation-Aware-Prototype-Learning-Approach-for-Few-Shot-Document-Level-Relation-Extraction/image-20241002162231121.png" alt="image-20241002162231121"></p>
</div>
<footer class="post-footer">
<div class="post-eof"></div>
</footer>
</article>
<article itemscope itemtype="http://schema.org/Article" class="post-block" lang="zh-CN">
<link itemprop="mainEntityOfPage" href="http://dawnzzz.github.io/2024/10/01/%E5%85%B3%E7%B3%BB%E6%8A%BD%E5%8F%96%E8%AE%BA%E6%96%87-Entity-Structure-Within-and-Throughout-Modeling-Mention-Dependencies-for-Document-Level-Relation-Extraction/">
<span hidden itemprop="author" itemscope itemtype="http://schema.org/Person">
<meta itemprop="image" content="/images/avatar.gif">
<meta itemprop="name" content="DawnZH">
<meta itemprop="description" content="个人博客站点">
</span>
<span hidden itemprop="publisher" itemscope itemtype="http://schema.org/Organization">
<meta itemprop="name" content="Dawn's Blogs">
</span>
<header class="post-header">
<h2 class="post-title" itemprop="name headline">
<a href="/2024/10/01/%E5%85%B3%E7%B3%BB%E6%8A%BD%E5%8F%96%E8%AE%BA%E6%96%87-Entity-Structure-Within-and-Throughout-Modeling-Mention-Dependencies-for-Document-Level-Relation-Extraction/" class="post-title-link" itemprop="url">关系抽取论文 Entity Structure Within and Throughout: Modeling Mention Dependencies for Document-Level Relation Extraction</a>
</h2>
<div class="post-meta">
<span class="post-meta-item">
<span class="post-meta-item-icon">
<i class="far fa-calendar"></i>
</span>
<span class="post-meta-item-text">发表于</span>
<time title="创建时间：2024-10-01 20:59:11 / 修改时间：21:55:52" itemprop="dateCreated datePublished" datetime="2024-10-01T20:59:11+08:00">2024-10-01</time>
</span>
<span class="post-meta-item">
<span class="post-meta-item-icon">
<i class="far fa-folder"></i>
</span>
<span class="post-meta-item-text">分类于</span>
<span itemprop="about" itemscope itemtype="http://schema.org/Thing">
<a href="/categories/%E5%85%B3%E7%B3%BB%E6%8A%BD%E5%8F%96%E8%AE%BA%E6%96%87/" itemprop="url" rel="index"><span itemprop="name">关系抽取论文</span></a>
</span>
</span>
</div>
</header>
<div class="post-body" itemprop="articleBody">
<blockquote>
<p>ntity Structure Within and Throughout: Modeling Mention Dependencies for Document-Level Relation Extraction</p>
<p>会议：AAAI</p>
<p>年份：2021</p>
<p>作者：Benfeng Xu, Quan Wang etc.</p>
<p>机构：School of Information Science and Technology, University of Science and Technology of China, Hefei, China</p>
<p>数据集：</p>
<ul>
<li>DocRED</li>
<li>CRD、GDA：生物医学领域</li>
</ul>
<p>贡献：</p>
<ul>
<li>将文档级文本中显示的各种<strong>提及依赖关系</strong>总结成一个统一的框架。通过在编码网络内部和整个编码网络中显式地合并这种结构，能够同时、交互式地执行上下文推理和结构推理，从而大大提高了关系提取任务的性能。</li>
<li>提出SSAN（Structured Self-Attention Network），通过结构性的指导，扩展标准的自注意力机制。</li>
</ul>
</blockquote>
<h1 id="Approach"><a href="#Approach" class="headerlink" title="Approach"></a>Approach</h1><h2 id="实体结构"><a href="#实体结构" class="headerlink" title="实体结构"></a>实体结构</h2><p>文档级关系抽取中的实体结构，两种结构都可以表示为True和False：</p>
<ul>
<li>共现（co-occurence）结构：一个句子中存在多个实体提及，表示句子内部的实体依赖关系。</li>
<li>协同（coreference）结构：多个实体提及指向同一实体，指向同一实体时需要被同时发现和推理。</li>
</ul>
<p><img src="/../images/%E5%85%B3%E7%B3%BB%E6%8A%BD%E5%8F%96%E8%AE%BA%E6%96%87-Entity-Structure-Within-and-Throughout-Modeling-Mention-Dependencies-for-Document-Level-Relation-Extraction/image-20241001213808222.png" alt="image-20241001213808222"></p>
<p>除了实体提及之间的依赖关系外，我们还进一步考虑了实体提及与其句子内非实体（None-Entity，NE）词之间的另一种依赖关系，把它表示为intraNE。对于其他句子间非实体词，我们假设不存在关键依赖，并将其归类为NA。</p>
<p>综上，实体结构为：intra+coref，intra+relate，inter+coref，inter+relate，intraNE，NA</p>
<h2 id="模型"><a href="#模型" class="headerlink" title="模型"></a>模型</h2><p><img src="/../images/%E5%85%B3%E7%B3%BB%E6%8A%BD%E5%8F%96%E8%AE%BA%E6%96%87-Entity-Structure-Within-and-Throughout-Modeling-Mention-Dependencies-for-Document-Level-Relation-Extraction/image-20241001213951334.png" alt="image-20241001213951334"></p>
<p>文章使用了引入结构性信息的Transformer结构，在计算注意力权重时引入结构信息，其中 s_i_j为上图中矩阵中的实体结构类型。</p>
<p><img src="/../images/%E5%85%B3%E7%B3%BB%E6%8A%BD%E5%8F%96%E8%AE%BA%E6%96%87-Entity-Structure-Within-and-Throughout-Modeling-Mention-Dependencies-for-Document-Level-Relation-Extraction/image-20241001214649255.png" alt="image-20241001214649255"></p>
<p><img src="/../images/%E5%85%B3%E7%B3%BB%E6%8A%BD%E5%8F%96%E8%AE%BA%E6%96%87-Entity-Structure-Within-and-Throughout-Modeling-Mention-Dependencies-for-Document-Level-Relation-Extraction/image-20241001214757373.png" alt="image-20241001214757373"></p>
<p>其中，转换模块（transformation）有两种方案，双仿射方案和线性分解方案：</p>
<p><img src="/../images/%E5%85%B3%E7%B3%BB%E6%8A%BD%E5%8F%96%E8%AE%BA%E6%96%87-Entity-Structure-Within-and-Throughout-Modeling-Mention-Dependencies-for-Document-Level-Relation-Extraction/image-20241001215131606.png" alt="image-20241001215131606"></p>
<p>在关系分类时，使用指向同一个实体的多个实体提及的平均池化作为实体表示，最后进行多分类。损失使用普通的交叉熵损失。</p>
<h1 id="实验结果"><a href="#实验结果" class="headerlink" title="实验结果"></a>实验结果</h1><p><img src="/../images/%E5%85%B3%E7%B3%BB%E6%8A%BD%E5%8F%96%E8%AE%BA%E6%96%87-Entity-Structure-Within-and-Throughout-Modeling-Mention-Dependencies-for-Document-Level-Relation-Extraction/image-20241001215440299.png" alt="image-20241001215440299"></p>
<p><img src="/../images/%E5%85%B3%E7%B3%BB%E6%8A%BD%E5%8F%96%E8%AE%BA%E6%96%87-Entity-Structure-Within-and-Throughout-Modeling-Mention-Dependencies-for-Document-Level-Relation-Extraction/image-20241001215454711.png" alt="image-20241001215454711"></p>
<p><img src="/../images/%E5%85%B3%E7%B3%BB%E6%8A%BD%E5%8F%96%E8%AE%BA%E6%96%87-Entity-Structure-Within-and-Throughout-Modeling-Mention-Dependencies-for-Document-Level-Relation-Extraction/image-20241001215510951.png" alt="image-20241001215510951"></p>
</div>
<footer class="post-footer">
<div class="post-eof"></div>
</footer>
</article>
<article itemscope itemtype="http://schema.org/Article" class="post-block" lang="zh-CN">
<link itemprop="mainEntityOfPage" href="http://dawnzzz.github.io/2024/05/23/%E5%A4%A7%E8%90%A5%E9%94%80%E5%B9%B3%E5%8F%B0%E9%A1%B9%E7%9B%AE-7-%E8%BF%94%E5%88%A9%E9%A2%86%E5%9F%9F%E4%B9%8B%E6%95%B0%E6%8D%AE%E5%BA%93%E8%AE%BE%E8%AE%A1-%E8%BF%94%E5%88%A9%E8%A1%8C%E4%B8%BA/">
<span hidden itemprop="author" itemscope itemtype="http://schema.org/Person">
<meta itemprop="image" content="/images/avatar.gif">
<meta itemprop="name" content="DawnZH">
<meta itemprop="description" content="个人博客站点">
</span>
<span hidden itemprop="publisher" itemscope itemtype="http://schema.org/Organization">
<meta itemprop="name" content="Dawn's Blogs">
</span>
<header class="post-header">
<h2 class="post-title" itemprop="name headline">
<a href="/2024/05/23/%E5%A4%A7%E8%90%A5%E9%94%80%E5%B9%B3%E5%8F%B0%E9%A1%B9%E7%9B%AE-7-%E8%BF%94%E5%88%A9%E9%A2%86%E5%9F%9F%E4%B9%8B%E6%95%B0%E6%8D%AE%E5%BA%93%E8%AE%BE%E8%AE%A1-%E8%BF%94%E5%88%A9%E8%A1%8C%E4%B8%BA/" class="post-title-link" itemprop="url">大营销平台项目 (7) 返利领域之数据库设计 返利行为</a>
</h2>
<div class="post-meta">
<span class="post-meta-item">
<span class="post-meta-item-icon">
<i class="far fa-calendar"></i>
</span>
<span class="post-meta-item-text">发表于</span>
<time title="创建时间：2024-05-23 15:32:22" itemprop="dateCreated datePublished" datetime="2024-05-23T15:32:22+08:00">2024-05-23</time>
</span>
<span class="post-meta-item">
<span class="post-meta-item-icon">
<i class="far fa-calendar-check"></i>
</span>
<span class="post-meta-item-text">更新于</span>
<time title="修改时间：2024-08-27 20:13:00" itemprop="dateModified" datetime="2024-08-27T20:13:00+08:00">2024-08-27</time>
</span>
<span class="post-meta-item">
<span class="post-meta-item-icon">
<i class="far fa-folder"></i>
</span>
<span class="post-meta-item-text">分类于</span>
<span itemprop="about" itemscope itemtype="http://schema.org/Thing">
<a href="/categories/%E5%A4%A7%E8%90%A5%E9%94%80%E5%B9%B3%E5%8F%B0%E9%A1%B9%E7%9B%AE/" itemprop="url" rel="index"><span itemprop="name">大营销平台项目</span></a>
</span>
</span>
</div>
</header>
<div class="post-body" itemprop="articleBody">
这是一篇加密文章，需要密码才能继续阅读
<!--noindex-->
<div class="post-button">
<a class="btn" href="/2024/05/23/%E5%A4%A7%E8%90%A5%E9%94%80%E5%B9%B3%E5%8F%B0%E9%A1%B9%E7%9B%AE-7-%E8%BF%94%E5%88%A9%E9%A2%86%E5%9F%9F%E4%B9%8B%E6%95%B0%E6%8D%AE%E5%BA%93%E8%AE%BE%E8%AE%A1-%E8%BF%94%E5%88%A9%E8%A1%8C%E4%B8%BA/#more" rel="contents">
阅读全文 »
</a>
</div>
<!--/noindex-->
</div>
<footer class="post-footer">
<div class="post-eof"></div>
</footer>
</article>
<article itemscope itemtype="http://schema.org/Article" class="post-block" lang="zh-CN">
<link itemprop="mainEntityOfPage" href="http://dawnzzz.github.io/2024/05/23/%E5%A4%A7%E8%90%A5%E9%94%80%E5%B9%B3%E5%8F%B0%E9%A1%B9%E7%9B%AE-6-%E5%A5%96%E5%93%81%E9%A2%86%E5%9F%9F%E4%B9%8B%E6%95%B0%E6%8D%AE%E5%BA%93%E8%AE%BE%E8%AE%A1-%E5%86%99%E5%85%A5%E4%B8%AD%E5%A5%96%E8%AE%B0%E5%BD%95/">
<span hidden itemprop="author" itemscope itemtype="http://schema.org/Person">
<meta itemprop="image" content="/images/avatar.gif">
<meta itemprop="name" content="DawnZH">
<meta itemprop="description" content="个人博客站点">
</span>
<span hidden itemprop="publisher" itemscope itemtype="http://schema.org/Organization">
<meta itemprop="name" content="Dawn's Blogs">
</span>
<header class="post-header">
<h2 class="post-title" itemprop="name headline">
<a href="/2024/05/23/%E5%A4%A7%E8%90%A5%E9%94%80%E5%B9%B3%E5%8F%B0%E9%A1%B9%E7%9B%AE-6-%E5%A5%96%E5%93%81%E9%A2%86%E5%9F%9F%E4%B9%8B%E6%95%B0%E6%8D%AE%E5%BA%93%E8%AE%BE%E8%AE%A1-%E5%86%99%E5%85%A5%E4%B8%AD%E5%A5%96%E8%AE%B0%E5%BD%95/" class="post-title-link" itemprop="url">大营销平台项目 (6) 奖品领域之数据库表设计 写入中奖记录</a>
</h2>
<div class="post-meta">
<span class="post-meta-item">
<span class="post-meta-item-icon">
<i class="far fa-calendar"></i>
</span>
<span class="post-meta-item-text">发表于</span>
<time title="创建时间：2024-05-23 15:32:17" itemprop="dateCreated datePublished" datetime="2024-05-23T15:32:17+08:00">2024-05-23</time>
</span>
<span class="post-meta-item">
<span class="post-meta-item-icon">
<i class="far fa-calendar-check"></i>
</span>
<span class="post-meta-item-text">更新于</span>
<time title="修改时间：2024-08-27 20:12:55" itemprop="dateModified" datetime="2024-08-27T20:12:55+08:00">2024-08-27</time>
</span>
<span class="post-meta-item">
<span class="post-meta-item-icon">
<i class="far fa-folder"></i>
</span>
<span class="post-meta-item-text">分类于</span>
<span itemprop="about" itemscope itemtype="http://schema.org/Thing">
<a href="/categories/%E5%A4%A7%E8%90%A5%E9%94%80%E5%B9%B3%E5%8F%B0%E9%A1%B9%E7%9B%AE/" itemprop="url" rel="index"><span itemprop="name">大营销平台项目</span></a>
</span>
</span>
</div>
</header>
<div class="post-body" itemprop="articleBody">
这是一篇加密文章，需要密码才能继续阅读
<!--noindex-->
<div class="post-button">
<a class="btn" href="/2024/05/23/%E5%A4%A7%E8%90%A5%E9%94%80%E5%B9%B3%E5%8F%B0%E9%A1%B9%E7%9B%AE-6-%E5%A5%96%E5%93%81%E9%A2%86%E5%9F%9F%E4%B9%8B%E6%95%B0%E6%8D%AE%E5%BA%93%E8%AE%BE%E8%AE%A1-%E5%86%99%E5%85%A5%E4%B8%AD%E5%A5%96%E8%AE%B0%E5%BD%95/#more" rel="contents">
阅读全文 »
</a>
</div>
<!--/noindex-->
</div>
<footer class="post-footer">
<div class="post-eof"></div>
</footer>
</article>
<article itemscope itemtype="http://schema.org/Article" class="post-block" lang="zh-CN">
<link itemprop="mainEntityOfPage" href="http://dawnzzz.github.io/2024/05/23/%E5%A4%A7%E8%90%A5%E9%94%80%E5%B9%B3%E5%8F%B0%E9%A1%B9%E7%9B%AE-5-%E6%B4%BB%E5%8A%A8%E9%A2%86%E5%9F%9F%E4%B9%8B%E5%88%9B%E5%BB%BAsku%E8%AE%A2%E5%8D%95-%E5%88%9B%E5%BB%BA%E6%8A%BD%E5%A5%96%E8%AE%A2%E5%8D%95/">
<span hidden itemprop="author" itemscope itemtype="http://schema.org/Person">
<meta itemprop="image" content="/images/avatar.gif">
<meta itemprop="name" content="DawnZH">
<meta itemprop="description" content="个人博客站点">
</span>
<span hidden itemprop="publisher" itemscope itemtype="http://schema.org/Organization">
<meta itemprop="name" content="Dawn's Blogs">
</span>
<header class="post-header">
<h2 class="post-title" itemprop="name headline">
<a href="/2024/05/23/%E5%A4%A7%E8%90%A5%E9%94%80%E5%B9%B3%E5%8F%B0%E9%A1%B9%E7%9B%AE-5-%E6%B4%BB%E5%8A%A8%E9%A2%86%E5%9F%9F%E4%B9%8B%E5%88%9B%E5%BB%BAsku%E8%AE%A2%E5%8D%95-%E5%88%9B%E5%BB%BA%E6%8A%BD%E5%A5%96%E8%AE%A2%E5%8D%95/" class="post-title-link" itemprop="url">大营销平台项目 (5) 活动领域之创建sku订单 创建抽奖订单</a>
</h2>
<div class="post-meta">
<span class="post-meta-item">
<span class="post-meta-item-icon">
<i class="far fa-calendar"></i>
</span>
<span class="post-meta-item-text">发表于</span>
<time title="创建时间：2024-05-23 15:32:06" itemprop="dateCreated datePublished" datetime="2024-05-23T15:32:06+08:00">2024-05-23</time>
</span>
<span class="post-meta-item">
<span class="post-meta-item-icon">
<i class="far fa-calendar-check"></i>
</span>
<span class="post-meta-item-text">更新于</span>
<time title="修改时间：2024-08-27 20:12:51" itemprop="dateModified" datetime="2024-08-27T20:12:51+08:00">2024-08-27</time>
</span>
<span class="post-meta-item">
<span class="post-meta-item-icon">
<i class="far fa-folder"></i>
</span>
<span class="post-meta-item-text">分类于</span>
<span itemprop="about" itemscope itemtype="http://schema.org/Thing">
<a href="/categories/%E5%A4%A7%E8%90%A5%E9%94%80%E5%B9%B3%E5%8F%B0%E9%A1%B9%E7%9B%AE/" itemprop="url" rel="index"><span itemprop="name">大营销平台项目</span></a>
</span>
</span>
</div>
</header>
<div class="post-body" itemprop="articleBody">
这是一篇加密文章，需要密码才能继续阅读
<!--noindex-->
<div class="post-button">
<a class="btn" href="/2024/05/23/%E5%A4%A7%E8%90%A5%E9%94%80%E5%B9%B3%E5%8F%B0%E9%A1%B9%E7%9B%AE-5-%E6%B4%BB%E5%8A%A8%E9%A2%86%E5%9F%9F%E4%B9%8B%E5%88%9B%E5%BB%BAsku%E8%AE%A2%E5%8D%95-%E5%88%9B%E5%BB%BA%E6%8A%BD%E5%A5%96%E8%AE%A2%E5%8D%95/#more" rel="contents">
阅读全文 »
</a>
</div>
<!--/noindex-->
</div>
<footer class="post-footer">
<div class="post-eof"></div>
</footer>
</article>
<article itemscope itemtype="http://schema.org/Article" class="post-block" lang="zh-CN">
<link itemprop="mainEntityOfPage" href="http://dawnzzz.github.io/2024/05/23/%E5%A4%A7%E8%90%A5%E9%94%80%E5%B9%B3%E5%8F%B0%E9%A1%B9%E7%9B%AE-4-%E6%B4%BB%E5%8A%A8%E9%A2%86%E5%9F%9F%E4%B9%8B%E6%95%B0%E6%8D%AE%E5%BA%93%E8%AE%BE%E8%AE%A1-%E6%B4%BB%E5%8A%A8%E8%A3%85%E9%85%8D/">
<span hidden itemprop="author" itemscope itemtype="http://schema.org/Person">
<meta itemprop="image" content="/images/avatar.gif">
<meta itemprop="name" content="DawnZH">
<meta itemprop="description" content="个人博客站点">
</span>
<span hidden itemprop="publisher" itemscope itemtype="http://schema.org/Organization">
<meta itemprop="name" content="Dawn's Blogs">
</span>
<header class="post-header">
<h2 class="post-title" itemprop="name headline">
<a href="/2024/05/23/%E5%A4%A7%E8%90%A5%E9%94%80%E5%B9%B3%E5%8F%B0%E9%A1%B9%E7%9B%AE-4-%E6%B4%BB%E5%8A%A8%E9%A2%86%E5%9F%9F%E4%B9%8B%E6%95%B0%E6%8D%AE%E5%BA%93%E8%AE%BE%E8%AE%A1-%E6%B4%BB%E5%8A%A8%E8%A3%85%E9%85%8D/" class="post-title-link" itemprop="url">大营销平台项目 (4) 活动领域之数据库设计 活动装配</a>
</h2>
<div class="post-meta">
<span class="post-meta-item">
<span class="post-meta-item-icon">
<i class="far fa-calendar"></i>
</span>
<span class="post-meta-item-text">发表于</span>
<time title="创建时间：2024-05-23 15:26:47" itemprop="dateCreated datePublished" datetime="2024-05-23T15:26:47+08:00">2024-05-23</time>
</span>
<span class="post-meta-item">
<span class="post-meta-item-icon">
<i class="far fa-calendar-check"></i>
</span>
<span class="post-meta-item-text">更新于</span>
<time title="修改时间：2024-08-27 20:12:48" itemprop="dateModified" datetime="2024-08-27T20:12:48+08:00">2024-08-27</time>
</span>
<span class="post-meta-item">
<span class="post-meta-item-icon">
<i class="far fa-folder"></i>
</span>
<span class="post-meta-item-text">分类于</span>
<span itemprop="about" itemscope itemtype="http://schema.org/Thing">
<a href="/categories/%E5%A4%A7%E8%90%A5%E9%94%80%E5%B9%B3%E5%8F%B0%E9%A1%B9%E7%9B%AE/" itemprop="url" rel="index"><span itemprop="name">大营销平台项目</span></a>
</span>
</span>
</div>
</header>
<div class="post-body" itemprop="articleBody">
这是一篇加密文章，需要密码才能继续阅读
<!--noindex-->
<div class="post-button">
<a class="btn" href="/2024/05/23/%E5%A4%A7%E8%90%A5%E9%94%80%E5%B9%B3%E5%8F%B0%E9%A1%B9%E7%9B%AE-4-%E6%B4%BB%E5%8A%A8%E9%A2%86%E5%9F%9F%E4%B9%8B%E6%95%B0%E6%8D%AE%E5%BA%93%E8%AE%BE%E8%AE%A1-%E6%B4%BB%E5%8A%A8%E8%A3%85%E9%85%8D/#more" rel="contents">
阅读全文 »
</a>
</div>
<!--/noindex-->
</div>
<footer class="post-footer">
<div class="post-eof"></div>
</footer>
</article>
<article itemscope itemtype="http://schema.org/Article" class="post-block" lang="zh-CN">
<link itemprop="mainEntityOfPage" href="http://dawnzzz.github.io/2024/05/23/%E5%A4%A7%E8%90%A5%E9%94%80%E5%B9%B3%E5%8F%B0%E9%A1%B9%E7%9B%AE-3-%E6%8A%BD%E5%A5%96%E7%AD%96%E7%95%A5%E9%A2%86%E5%9F%9F%E4%B9%8B%E8%B4%A3%E4%BB%BB%E9%93%BE-%E8%A7%84%E5%88%99%E6%A0%91/">
<span hidden itemprop="author" itemscope itemtype="http://schema.org/Person">
<meta itemprop="image" content="/images/avatar.gif">
<meta itemprop="name" content="DawnZH">
<meta itemprop="description" content="个人博客站点">
</span>
<span hidden itemprop="publisher" itemscope itemtype="http://schema.org/Organization">
<meta itemprop="name" content="Dawn's Blogs">
</span>
<header class="post-header">
<h2 class="post-title" itemprop="name headline">
<a href="/2024/05/23/%E5%A4%A7%E8%90%A5%E9%94%80%E5%B9%B3%E5%8F%B0%E9%A1%B9%E7%9B%AE-3-%E6%8A%BD%E5%A5%96%E7%AD%96%E7%95%A5%E9%A2%86%E5%9F%9F%E4%B9%8B%E8%B4%A3%E4%BB%BB%E9%93%BE-%E8%A7%84%E5%88%99%E6%A0%91/" class="post-title-link" itemprop="url">大营销平台项目 (3) 抽奖策略领域之责任链 规则树</a>
</h2>
<div class="post-meta">
<span class="post-meta-item">
<span class="post-meta-item-icon">
<i class="far fa-calendar"></i>
</span>
<span class="post-meta-item-text">发表于</span>
<time title="创建时间：2024-05-23 15:18:39" itemprop="dateCreated datePublished" datetime="2024-05-23T15:18:39+08:00">2024-05-23</time>
</span>
<span class="post-meta-item">
<span class="post-meta-item-icon">
<i class="far fa-calendar-check"></i>
</span>
<span class="post-meta-item-text">更新于</span>
<time title="修改时间：2024-08-27 20:12:44" itemprop="dateModified" datetime="2024-08-27T20:12:44+08:00">2024-08-27</time>
</span>
<span class="post-meta-item">
<span class="post-meta-item-icon">
<i class="far fa-folder"></i>
</span>
<span class="post-meta-item-text">分类于</span>
<span itemprop="about" itemscope itemtype="http://schema.org/Thing">
<a href="/categories/%E5%A4%A7%E8%90%A5%E9%94%80%E5%B9%B3%E5%8F%B0%E9%A1%B9%E7%9B%AE/" itemprop="url" rel="index"><span itemprop="name">大营销平台项目</span></a>
</span>
</span>
</div>
</header>
<div class="post-body" itemprop="articleBody">
这是一篇加密文章，需要密码才能继续阅读
<!--noindex-->
<div class="post-button">
<a class="btn" href="/2024/05/23/%E5%A4%A7%E8%90%A5%E9%94%80%E5%B9%B3%E5%8F%B0%E9%A1%B9%E7%9B%AE-3-%E6%8A%BD%E5%A5%96%E7%AD%96%E7%95%A5%E9%A2%86%E5%9F%9F%E4%B9%8B%E8%B4%A3%E4%BB%BB%E9%93%BE-%E8%A7%84%E5%88%99%E6%A0%91/#more" rel="contents">
阅读全文 »
</a>
</div>
<!--/noindex-->
</div>
<footer class="post-footer">
<div class="post-eof"></div>
</footer>
</article>
<article itemscope itemtype="http://schema.org/Article" class="post-block" lang="zh-CN">
<link itemprop="mainEntityOfPage" href="http://dawnzzz.github.io/2024/05/23/%E5%A4%A7%E8%90%A5%E9%94%80%E5%B9%B3%E5%8F%B0%E9%A1%B9%E7%9B%AE-2-%E6%8A%BD%E5%A5%96%E7%AD%96%E7%95%A5%E9%A2%86%E5%9F%9F%E4%B9%8B%E6%95%B0%E6%8D%AE%E5%BA%93%E8%AE%BE%E8%AE%A1-%E7%AD%96%E7%95%A5%E8%A3%85%E9%85%8D/">
<span hidden itemprop="author" itemscope itemtype="http://schema.org/Person">
<meta itemprop="image" content="/images/avatar.gif">
<meta itemprop="name" content="DawnZH">
<meta itemprop="description" content="个人博客站点">
</span>
<span hidden itemprop="publisher" itemscope itemtype="http://schema.org/Organization">
<meta itemprop="name" content="Dawn's Blogs">
</span>
<header class="post-header">
<h2 class="post-title" itemprop="name headline">
<a href="/2024/05/23/%E5%A4%A7%E8%90%A5%E9%94%80%E5%B9%B3%E5%8F%B0%E9%A1%B9%E7%9B%AE-2-%E6%8A%BD%E5%A5%96%E7%AD%96%E7%95%A5%E9%A2%86%E5%9F%9F%E4%B9%8B%E6%95%B0%E6%8D%AE%E5%BA%93%E8%AE%BE%E8%AE%A1-%E7%AD%96%E7%95%A5%E8%A3%85%E9%85%8D/" class="post-title-link" itemprop="url">大营销平台项目 (2) 抽奖策略领域之数据库设计 策略装配</a>
</h2>
<div class="post-meta">
<span class="post-meta-item">
<span class="post-meta-item-icon">
<i class="far fa-calendar"></i>
</span>
<span class="post-meta-item-text">发表于</span>
<time title="创建时间：2024-05-23 15:14:12" itemprop="dateCreated datePublished" datetime="2024-05-23T15:14:12+08:00">2024-05-23</time>
</span>
<span class="post-meta-item">
<span class="post-meta-item-icon">
<i class="far fa-calendar-check"></i>
</span>
<span class="post-meta-item-text">更新于</span>
<time title="修改时间：2024-08-27 20:12:40" itemprop="dateModified" datetime="2024-08-27T20:12:40+08:00">2024-08-27</time>
</span>
<span class="post-meta-item">
<span class="post-meta-item-icon">
<i class="far fa-folder"></i>
</span>
<span class="post-meta-item-text">分类于</span>
<span itemprop="about" itemscope itemtype="http://schema.org/Thing">
<a href="/categories/%E5%A4%A7%E8%90%A5%E9%94%80%E5%B9%B3%E5%8F%B0%E9%A1%B9%E7%9B%AE/" itemprop="url" rel="index"><span itemprop="name">大营销平台项目</span></a>
</span>
</span>
</div>
</header>
<div class="post-body" itemprop="articleBody">
这是一篇加密文章，需要密码才能继续阅读
<!--noindex-->
<div class="post-button">
<a class="btn" href="/2024/05/23/%E5%A4%A7%E8%90%A5%E9%94%80%E5%B9%B3%E5%8F%B0%E9%A1%B9%E7%9B%AE-2-%E6%8A%BD%E5%A5%96%E7%AD%96%E7%95%A5%E9%A2%86%E5%9F%9F%E4%B9%8B%E6%95%B0%E6%8D%AE%E5%BA%93%E8%AE%BE%E8%AE%A1-%E7%AD%96%E7%95%A5%E8%A3%85%E9%85%8D/#more" rel="contents">
阅读全文 »
</a>
</div>
<!--/noindex-->
</div>
<footer class="post-footer">
<div class="post-eof"></div>
</footer>
</article>
<article itemscope itemtype="http://schema.org/Article" class="post-block" lang="zh-CN">
<link itemprop="mainEntityOfPage" href="http://dawnzzz.github.io/2024/05/23/%E5%A4%A7%E8%90%A5%E9%94%80%E5%B9%B3%E5%8F%B0%E9%A1%B9%E7%9B%AE-1-DDD%E9%A2%86%E5%9F%9F%E9%A9%B1%E5%8A%A8%E8%AE%BE%E8%AE%A1/">
<span hidden itemprop="author" itemscope itemtype="http://schema.org/Person">
<meta itemprop="image" content="/images/avatar.gif">