词性标注分类体系：https://blog.csdn.net/qq_32023541/article/details/84632829
-
原始数据：ontonotes-release-5.0/
-
处理程序1：util/process_ontonotes_data_to_word.py
说明：把原始ontonotes数据用BasicTokenizer切分，中文一个字为一个单位，英文一个单词为一个单位 -
处理程序2：util/replace_tag.sh
说明：把步骤2得到的数据中的不在词性标注分类的tag去掉或替换 -
处理程序3：util/split_ontonotes_data.py
说明：把步骤3处理过的数据，分割为train/dev/test三份 -
处理程序4：util/process_ontonotes_word_to_wordpiece.py
说明：把原始ontonotes数据处理成标准NER格式的数据，用WordpieceTokenizer切分，适合bert使用。 注意：Bert给定中文词表是不包括大写字母(A-Z)的，所以必须得do_lower_case. 英文词表包括大写字母，不用do_lower_case.
cd bert_pos/
train: sh run_train.sh
export: sh run_export.sh
ontonotes测试集效果（ontonotes训练集训练）:
the macro F1 is: 0.856473
the micro F1 is: 0.942755
================================================================================
tp = 204922 tn = 1172907
================================================================================
tag tp fp fn Precsion Recall F1
B-NN 26627 1678 1578 0.940717 0.944052 0.942382
I-NN 28076 1421 1607 0.951826 0.945861 0.948834
B-PU 21446 72 52 0.996654 0.997581 0.997117
I-PU 389 22 71 0.946472 0.845652 0.893226
B-VV 18466 1167 1175 0.940559 0.940176 0.940368
I-VV 13366 1076 1138 0.925495 0.921539 0.923513
B-AD 12497 614 773 0.953169 0.941748 0.947424
I-AD 5995 616 670 0.906822 0.899475 0.903133
B-NR 6243 292 182 0.955318 0.971673 0.963426
I-NR 10729 708 259 0.938096 0.976429 0.956878
B-PN 5981 166 203 0.972995 0.967173 0.970075
I-PN 2058 96 120 0.955432 0.944904 0.950139
B-P 4382 191 233 0.958233 0.949512 0.953853
I-P 752 37 38 0.953105 0.951899 0.952502
B-CD 3777 189 165 0.952345 0.958143 0.955235
I-CD 2096 112 84 0.949275 0.961468 0.955333
B-DEG 3896 192 332 0.953033 0.921476 0.936989
I-DEG 0 0 0 1.000000 1.000000 1.000000
B-M 3321 160 176 0.954036 0.949671 0.951849
I-M 147 18 39 0.890909 0.790323 0.837607
B-JJ 2344 492 692 0.826516 0.772069 0.798365
I-JJ 1881 435 526 0.812176 0.781471 0.796528
B-DEC 2539 358 226 0.876424 0.918264 0.896856
I-DEC 0 0 1 1.000000 0.000000 0.000000
B-DT 2603 183 116 0.934314 0.957337 0.945686
I-DT 1327 107 93 0.925384 0.934507 0.929923
B-VC 2346 102 107 0.958333 0.956380 0.957356
I-VC 0 0 2 1.000000 0.000000 0.000000
B-VA 2104 474 373 0.816137 0.849415 0.832443
I-VA 1367 432 340 0.759867 0.800820 0.779806
B-NT 2056 100 90 0.953618 0.958062 0.955834
I-NT 3012 133 92 0.957711 0.970361 0.963994
B-LC 1776 101 84 0.946191 0.954839 0.950495
I-LC 472 31 36 0.938370 0.929134 0.933729
B-SP 2031 137 141 0.936808 0.935083 0.935945
I-SP 58 2 5 0.966667 0.920635 0.943089
B-AS 1558 111 66 0.933493 0.959360 0.946250
I-AS 0 0 0 1.000000 1.000000 1.000000
B-CC 1387 98 82 0.934007 0.944180 0.939066
I-CC 206 18 47 0.919643 0.814229 0.863732
B-VE 1138 83 53 0.932023 0.955500 0.943615
I-VE 155 13 20 0.922619 0.885714 0.903790
B-IJ 1281 49 40 0.963158 0.969720 0.966428
I-IJ 81 9 45 0.900000 0.642857 0.750000
B-OD 281 30 57 0.903537 0.831361 0.865948
I-OD 259 7 37 0.973684 0.875000 0.921708
B-CS 348 16 21 0.956044 0.943089 0.949523
I-CS 337 19 24 0.946629 0.933518 0.940028
B-MSP 303 27 31 0.918182 0.907186 0.912651
I-MSP 0 0 0 1.000000 1.000000 1.000000
B-DEV 284 12 27 0.959459 0.913183 0.935750
I-DEV 0 0 0 1.000000 1.000000 1.000000
B-BA 245 2 2 0.991903 0.991903 0.991903
I-BA 0 0 0 1.000000 1.000000 1.000000
B-ETC 162 1 1 0.993865 0.993865 0.993865
I-ETC 19 1 0 0.950000 1.000000 0.974359
B-SB 159 5 10 0.969512 0.940828 0.954955
I-SB 0 0 0 1.000000 1.000000 1.000000
B-DER 90 10 13 0.900000 0.873786 0.886700
I-DER 0 0 0 1.000000 1.000000 1.000000
B-LB 73 9 7 0.890244 0.912500 0.901235
I-LB 0 0 0 1.000000 1.000000 1.000000
B-URL 18 0 2 1.000000 0.900000 0.947368
I-URL 378 8 20 0.979275 0.949749 0.964286
B-FW 0 0 11 1.000000 0.000000 0.000000
I-FW 0 0 1 1.000000 0.000000 0.000000
B-ON 0 0 5 1.000000 0.000000 0.000000
I-ON 0 1 2 0.000000 0.000000 0.000000
B-X 0 0 0 1.000000 1.000000 1.000000
I-X 0 0 0 1.000000 1.000000 1.000000
================================================================================
推理sdk
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent