Amanpreet Singh commited on
Commit
4607ab7
1 Parent(s): d22ddbe

Initial commit from biomedical workshop

Browse files
Files changed (1) hide show
  1. bioasq_task_b.py +596 -0
bioasq_task_b.py ADDED
@@ -0,0 +1,596 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # coding=utf-8
2
+ # Copyright 2022 The HuggingFace Datasets Authors and the current dataset script contributor.
3
+ #
4
+ # Licensed under the Apache License, Version 2.0 (the "License");
5
+ # you may not use this file except in compliance with the License.
6
+ # You may obtain a copy of the License at
7
+ #
8
+ # http://www.apache.org/licenses/LICENSE-2.0
9
+ #
10
+ # Unless required by applicable law or agreed to in writing, software
11
+ # distributed under the License is distributed on an "AS IS" BASIS,
12
+ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13
+ # See the License for the specific language governing permissions and
14
+ # limitations under the License.
15
+ """
16
+ BioASQ Task B On Biomedical Semantic QA (Involves IR, QA, Summarization qnd
17
+ More). This task uses benchmark datasets containing development and test
18
+ questions, in English, along with gold standard (reference) answers constructed
19
+ by a team of biomedical experts. The participants have to respond with relevant
20
+ concepts, articles, snippets and RDF triples, from designated resources, as well
21
+ as exact and 'ideal' answers.
22
+
23
+ Fore more information about the challenge, the organisers and the relevant
24
+ publications please visit: http://bioasq.org/
25
+ """
26
+ import glob
27
+ import json
28
+ import os
29
+ import re
30
+
31
+ import datasets
32
+ from utils import schemas
33
+ from utils.configs import BigBioConfig
34
+ from utils.constants import Tasks
35
+
36
+ _CITATION = """\
37
+ @article{tsatsaronis2015overview,
38
+ title = {
39
+ An overview of the BIOASQ large-scale biomedical semantic indexing and
40
+ question answering competition
41
+ },
42
+ author = {
43
+ Tsatsaronis, George and Balikas, Georgios and Malakasiotis, Prodromos
44
+ and Partalas, Ioannis and Zschunke, Matthias and Alvers, Michael R and
45
+ Weissenborn, Dirk and Krithara, Anastasia and Petridis, Sergios and
46
+ Polychronopoulos, Dimitris and others
47
+ },
48
+ year = 2015,
49
+ journal = {BMC bioinformatics},
50
+ publisher = {BioMed Central Ltd},
51
+ volume = 16,
52
+ number = 1,
53
+ pages = 138
54
+ }
55
+ """
56
+
57
+ _DATASETNAME = "bioasq"
58
+
59
+ _BIOASQ_10B_DESCRIPTION = """\
60
+ The data are intended to be used as training and development data for BioASQ
61
+ 10, which will take place during 2022. There is one file containing the data:
62
+ - training10b.json
63
+
64
+ The file contains the data of the first nine editions of the challenge: 4234
65
+ questions [1] with their relevant documents, snippets, concepts and RDF
66
+ triples, exact and ideal answers.
67
+
68
+ Differences with BioASQ-training9b.json
69
+ - 492 new questions added from BioASQ9
70
+ - The question with id 56c1f01eef6e394741000046 had identical body with
71
+ 602498cb1cb411341a00009e. All relevant elements from both questions
72
+ are available in the merged question with id 602498cb1cb411341a00009e.
73
+ - The question with id 5c7039207c78d69471000065 had identical body with
74
+ 601c317a1cb411341a000014. All relevant elements from both questions
75
+ are available in the merged question with id 601c317a1cb411341a000014.
76
+ - The question with id 5e4b540b6d0a27794100001c had identical body with
77
+ 602828b11cb411341a0000fc. All relevant elements from both questions
78
+ are available in the merged question with id 602828b11cb411341a0000fc.
79
+ - The question with id 5fdb42fba43ad31278000027 had identical body with
80
+ 5d35eb01b3a638076300000f. All relevant elements from both questions
81
+ are available in the merged question with id 5d35eb01b3a638076300000f.
82
+ - The question with id 601d76311cb411341a000045 had identical body with
83
+ 6060732b94d57fd87900003d. All relevant elements from both questions
84
+ are available in the merged question with id 6060732b94d57fd87900003d.
85
+
86
+ [1] 4234 questions : 1252 factoid, 1148 yesno, 1018 summary, 816 list
87
+ """
88
+
89
+ _BIOASQ_9B_DESCRIPTION = """\
90
+ The data are intended to be used as training and development data for BioASQ 9,
91
+ which will take place during 2021. There is one file containing the data:
92
+ - training9b.json
93
+
94
+ The file contains the data of the first seven editions of the challenge: 3742
95
+ questions [1] with their relevant documents, snippets, concepts and RDF triples,
96
+ exact and ideal answers.
97
+
98
+ Differences with BioASQ-training8b.json
99
+ - 499 new questions added from BioASQ8
100
+ - The question with id 5e30e689fbd6abf43b00003a had identical body with
101
+ 5880e417713cbdfd3d000001. All relevant elements from both questions
102
+ are available in the merged question with id 5880e417713cbdfd3d000001.
103
+
104
+ [1] 3742 questions : 1091 factoid, 1033 yesno, 899 summary, 719 list
105
+ """
106
+
107
+ _BIOASQ_8B_DESCRIPTION = """\
108
+ The data are intended to be used as training and development data for BioASQ 8,
109
+ which will take place during 2020. There is one file containing the data:
110
+ - training8b.json
111
+
112
+ The file contains the data of the first seven editions of the challenge: 3243
113
+ questions [1] with their relevant documents, snippets, concepts and RDF triples,
114
+ exact and ideal answers.
115
+
116
+ Differences with BioASQ-training7b.json
117
+ - 500 new questions added from BioASQ7
118
+ - 4 questions were removed
119
+ - The question with id 5717fb557de986d80d000009 had identical body with
120
+ 571e06447de986d80d000016. All relevant elements from both questions
121
+ are available in the merged question with id 571e06447de986d80d000016.
122
+ - The question with id 5c589ddb86df2b917400000b had identical body with
123
+ 5c6b7a9e7c78d69471000029. All relevant elements from both questions
124
+ are available in the merged question with id 5c6b7a9e7c78d69471000029.
125
+ - The question with id 52ffb5d12059c6d71c00007c had identical body with
126
+ 52e7870a98d023950500001a. All relevant elements from both questions
127
+ are available in the merged question with id 52e7870a98d023950500001a.
128
+ - The question with id 53359338d6d3ac6a3400004f had identical body with
129
+ 589a246878275d0c4a000030. All relevant elements from both questions
130
+ are available in the merged question with id 589a246878275d0c4a000030.
131
+
132
+ **** UPDATE 25/02/2020 *****
133
+ The previous version of the dataset contained an inconsistency on question with
134
+ id "5c9904eaecadf2e73f00002e", where the "ideal_answer" field was missing.
135
+ This has been fixed.
136
+ """
137
+
138
+ _BIOASQ_7B_DESCRIPTION = """\
139
+ The data are intended to be used as training and development data for BioASQ 7,
140
+ which will take place during 2019. There is one file containing the data:
141
+ - BioASQ-trainingDataset7b.json
142
+
143
+ The file contains the data of the first six editions of the challenge: 2747
144
+ questions [1] with their relevant documents, snippets, concepts and RDF triples,
145
+ exact and ideal answers.
146
+
147
+ Differences with BioASQ-trainingDataset6b.json
148
+ - 500 new questions added from BioASQ6
149
+ - 4 questions were removed
150
+ - The question with id 569ed752ceceede94d000004 had identical body with
151
+ a new question from BioASQ6. All relevant elements from both questions
152
+ are available in the merged question with id 5abd31e0fcf456587200002c
153
+ - 3 questions were removed as incomplete: 54d643023706e89528000007,
154
+ 532819afd6d3ac6a3400000f, 517545168ed59a060a00002b
155
+ - 4 questions were revised for various confusions that have been identified
156
+ - In 2 questions the ideal answer has been revised :
157
+ 51406e6223fec90375000009, 5172f8118ed59a060a000019
158
+ - In 4 questions the snippets and documents list has been revised :
159
+ 51406e6223fec90375000009, 5172f8118ed59a060a000019,
160
+ 51593dc8d24251bc05000099, 5158a5b8d24251bc05000097
161
+ - In 198 questions the documents list has updated with missing
162
+ documents from the relevant snippets list. [2]
163
+
164
+ [1] 2747 questions : 779 factoid, 745 yesno, 667 summary, 556 list
165
+ [2] 55031181e9bde69634000014, 51406e6223fec90375000009, 54d643023706e89528000007,
166
+ 52bf1b0a03868f1b06000009, 52bf19c503868f1b06000001, 51593dc8d24251bc05000099,
167
+ 530a5117970c65fa6b000007, 553a8d78f321868558000003, 531a3fe3b166e2b806000038,
168
+ 532819afd6d3ac6a3400000f, 5158a5b8d24251bc05000097, 553653a5bc4f83e828000007,
169
+ 535d2cf09a4572de6f000004, 53386282d6d3ac6a3400005a, 517a8ce98ed59a060a000045,
170
+ 55391ce8bc4f83e828000018, 5547d700f35db75526000007, 5713bf261174fb1755000011,
171
+ 6f15c5a2ac5ed1459000012, 52b2e498f828ad283c000010, 570a7594cf1c325851000026,
172
+ 530cefaaad0bf1360c000012, 530f685c329f5fcf1e000002, 550c4011a103b78016000009,
173
+ 552faababc4f83e828000005, 54cf48acf693c3b16b00000b, 550313aae9bde6963400001f,
174
+ 551177626a8cde6b72000005, 54eded8c94afd6150400000c, 550c3754a103b78016000007,
175
+ 56f555b609dd18d46b000007, 54c26e29f693c3b16b000003, 54da0c524b1fd0d33c00000b,
176
+ 52bf1d3c03868f1b0600000d, 5343bdd6aeec6fbd07000001, 52cb9b9b03868f1b0600002d,
177
+ 55423875ec76f5e50c000002, 571366ba1174fb1755000005, 56c4d14ab04e159d0e000003,
178
+ 550c44d1a103b7801600000a, 5547a01cf35db75526000005, 55422640ccca0ce74b000004,
179
+ 54ecb66d445c3b5a5f000002, 553656c4bc4f83e828000009, 5172f8118ed59a060a000019,
180
+ 513711055274a5fb0700000e, 54d892ee014675820d000005, 52e6c92598d0239505000019,
181
+ 5353aedb288f4dae47000006, 52bf1f1303868f1b06000014, 5519113b622b19434500000f,
182
+ 52b2f1724003448f5500000b, 5525317687ecba3764000007, 554a0cadf35db7552600000f,
183
+ 55152bd246478f2f2c000002, 516c3960298dcd4e51000073, 571e417bbb137a4b0c00000a,
184
+ 551910d3622b194345000008, 54dc8ed6c0bb8dce23000002, 511a4ec01159fa8212000004,
185
+ 54d8ea2c4b1fd0d33c000002, 5148e1d6d24251bc0500003a, 515dbb3b298dcd4e51000018,
186
+ 56f7c15a09dd18d46b000012, 51475d5cd24251bc0500001b, 54db7c4ac0bb8dce23000001,
187
+ 57152ebbcb4ef8864c000002, 57134d511174fb1755000002, 55149f156a8cde6b72000013,
188
+ 56bcd422d36b5da378000005, 54ede5c394afd61504000006, 517545168ed59a060a00002b,
189
+ 5710ed19a5ed216440000003, 53442472aeec6fbd07000008, 55088e412e93f0133a000001,
190
+ 54d762653706e89528000014, 550aef0ec2af5d5b7000000a, 552435602c8b63434a000009,
191
+ 552446612c8b63434a00000c, 54d901ec4b1fd0d33c000006, 54cf45e7f693c3b16b00000a,
192
+ 52fc8b772059c6d71c00006e, 5314d05adae131f84700000d, 5512c91b6a8cde6b7200000b,
193
+ 56c5a7605795f9a73e000002, 55030a6ce9bde6963400000f, 553fac39c6a5098552000001,
194
+ 531a3a58b166e2b806000037, 5509bd6a1180f13250000002, 54f9c40ddd3fc62544000001,
195
+ 553c8fd1f32186855800000a, 56bce51cd36b5da37800000a, 550316a6e9bde69634000029,
196
+ 55031286e9bde6963400001b, 536e46f27d100faa09000012, 5502abd1e9bde69634000008,
197
+ 551af9106b348bb82c000002, 54edeb4394afd6150400000b, 5717cdd2070aa3d072000001,
198
+ 56c5ade15795f9a73e000003, 531464a6e3eabad021000014, 58a0d87a78275d0c4a000053,
199
+ 58a3160d60087bc10a00000a, 58a5d54860087bc10a000025, 58a0da5278275d0c4a000054,
200
+ 58a3264e60087bc10a00000d, 589c8ef878275d0c4a000042, 58a3428d60087bc10a00001b,
201
+ 58a3196360087bc10a00000b, 58a341eb60087bc10a000018, 58a3275960087bc10a00000f,
202
+ 58a342e760087bc10a00001c, 58bd645702b8c60953000010, 58bc8e5002b8c60953000006,
203
+ 58bc8e7a02b8c60953000007, 58a1da4e78275d0c4a000059, 58bcb83d02b8c6095300000f,
204
+ 58bc9a5002b8c60953000008, 589dee3778275d0c4a000050, 58a32efe60087bc10a000013,
205
+ 58a327bf60087bc10a000011, 58bca08702b8c6095300000a, 58bc9dbb02b8c60953000009,
206
+ 58c99fcc02b8c60953000029, 58bca2f302b8c6095300000c, 58cbf1f402b8c60953000036,
207
+ 58cdb41302b8c60953000042, 58cdb80302b8c60953000043, 58cdbaf302b8c60953000044,
208
+ 58cb305c02b8c60953000032, 58caf86f02b8c60953000030, 58c1b2f702b8c6095300001e,
209
+ 58bde18b02b8c60953000014, 58eb7898eda5a57672000006, 58caf88c02b8c60953000031,
210
+ 58e11bf76fddd3e83e00000c, 58cdbbd102b8c60953000045, 58df779d6fddd3e83e000001,
211
+ 58dbb4f08acda3452900001a, 58dbb8968acda3452900001b, 58add7699ef3c34033000009,
212
+ 58dbbbf08acda3452900001d, 58dbba438acda3452900001c, 58dd2cb08acda34529000029,
213
+ 58eb9542eda5a57672000007, 58f3ca5c70f9fc6f0f00000d, 58e9e7aa3e8b6dc87c00000d,
214
+ 58e3d9ab3e8b6dc87c000002, 58eb4ce7eda5a57672000004, 58f3c8f470f9fc6f0f00000c,
215
+ 58f3c62970f9fc6f0f00000b, 58adca6d9ef3c34033000007, 58f4b3ee70f9fc6f0f000013,
216
+ 593ff22b70f9fc6f0f000023, 5a679875b750ff4455000004, 5a774585faa1ab7d2e000005,
217
+ 5a6f7245b750ff4455000050, 5a787544faa1ab7d2e00000b, 5a74d9980384be9551000008,
218
+ 5a6a02a3b750ff4455000021, 5a6e47b1b750ff4455000049, 5a87124561bb38fb24000001,
219
+ 5a6e42f1b750ff4455000046, 5a8b1264fcd1d6a10c00001d, 5a981e66fcd1d6a10c00002f,
220
+ 5a8718c861bb38fb24000008, 5a7615af83b0d9ea6600001f, 5a87140a61bb38fb24000003,
221
+ 5a77072c9e632bc06600000a, 5a897601fcd1d6a10c000008, 5a871a6861bb38fb24000009,
222
+ 5a74e9ad0384be955100000a, 5a79d25dfaa1ab7d2e00000f, 5a6900ebb750ff445500001d,
223
+ 5a87145861bb38fb24000004, 5a871b8d61bb38fb2400000a, 5a897a06fcd1d6a10c00000b,
224
+ 5a8dc6b4fcd1d6a10c000026, 5a8712af61bb38fb24000002, 5a8714e261bb38fb24000005,
225
+ 5aa304f1d6d6b54f79000004, 5a981bcffcd1d6a10c00002d, 5aa3fa73d6d6b54f79000008,
226
+ 5aa55b45d6d6b54f7900000d, 5a981dd0fcd1d6a10c00002e, 5a9700adfcd1d6a10c00002c,
227
+ 5a9d8ffe1d1251d03b000022, 5a96c74cfcd1d6a10c000029, 5aa50086d6d6b54f7900000c,
228
+ 5a95765bfcd1d6a10c000028, 5a96f40cfcd1d6a10c00002b, 5ab144fefcf4565872000012,
229
+ 5aa67b4fd6d6b54f7900000f, 5abd5a62fcf4565872000031, 5abbe429fcf456587200001c,
230
+ 5aaef38dfcf456587200000f, 5abce6acfcf4565872000022, 5aae6499fcf456587200000c
231
+ """
232
+
233
+ _BIOASQ_6B_DESCRIPTION = """\
234
+ The data are intended to be used as training and development data for BioASQ 6,
235
+ which will take place during 2018. There is one file containing the data:
236
+ - BioASQ-trainingDataset6b.json
237
+
238
+ Differences with BioASQ-trainingDataset5b.json
239
+ - 500 new questions added from BioASQ5
240
+ - 48 pairs of questions with identical bodies have been merged into one
241
+ question having only one question-id, but all the documents, snippets,
242
+ concepts, RDF triples and answers of both questions of the pair.
243
+ - This normalization lead to the removal of 48 deprecated question
244
+ ids [2] from the dataset and to the update of the 48 remaining
245
+ questions [3].
246
+ - In cases where a pair of questions with identical bodies had some
247
+ inconsistency (e.g. different question type), the inconsistency has
248
+ been solved merging the pair manually consulting the BioASQ expert team.
249
+ - 12 questions were revised for various confusions that have been
250
+ identified
251
+ - In 8 questions the question type has been changed to better suit to
252
+ the question body. The change of type lead to corresponding changes
253
+ in exact answers existence and format : 54fc4e2e6ea36a810c000003,
254
+ 530b01a6970c65fa6b000008, 530cf54dab4de4de0c000009,
255
+ 531b2fc3b166e2b80600003c, 532819afd6d3ac6a3400000f,
256
+ 532aad53d6d3ac6a34000010, 5710ade4cf1c32585100002c,
257
+ 52f65f372059c6d71c000027
258
+ - In 6 questions the ideal answer has been revised :
259
+ 532aad53d6d3ac6a34000010, 5710ade4cf1c32585100002c,
260
+ 53147b52e3eabad021000015, 5147c8a6d24251bc05000027,
261
+ 5509bd6a1180f13250000002, 58bbb71f22d3005309000016
262
+ - In 5 questions the exact answer has been revised :
263
+ 5314bd7ddae131f847000006, 53130a77e3eabad02100000f,
264
+ 53148a07dae131f847000002, 53147b52e3eabad021000015,
265
+ 5147c8a6d24251bc05000027
266
+ - In 2 questions the question body has been revised :
267
+ 52f65f372059c6d71c000027, 5503145ee9bde69634000022
268
+ - In lists of ideal answers, documents, snippets, concepts and RDF triples
269
+ any duplicate identical elements have been removed.
270
+ - Ideal answers in format of one string have been converted to a list with
271
+ one element for consistency with cases where more than one golden ideal
272
+ answers are available. (i.e. "ideal_ans1" converted to ["ideal_ans1"])
273
+ - For yesno questions: All exact answers have been normalized to "yes" or
274
+ "no" (replacing "Yes", "YES" and "No")
275
+ - For factoid questions: The format of the exact answer was normalized to a
276
+ list of strings for each question, representing a set of synonyms
277
+ answering the question (i.e. [`ans1`, `syn11`, ... ]).
278
+ - For list questions: The format of the exact answer was normalized to a
279
+ list of lists. Each internal list represents one element of the answer
280
+ as a set of synonyms
281
+ (i.e. [[`ans1`, `syn11`, `syn12`], [`ans2`], [`ans3`, `syn31`] ...]).
282
+ - Empty elements, e.g. empty lists of documents have been removed.
283
+
284
+ [1] 2251 questions : 619 factoid, 616 yesno, 531 summary, 485 list
285
+ [2] The 48 deprecated question ids are : 52f8b2902059c6d71c000053,
286
+ 52f11bf22059c6d71c000005, 52f77edb2059c6d71c000028, 52ed795098d0239505000032,
287
+ 56d1a9baab2fed4a47000002, 52f7d3472059c6d71c00002f, 52fbe2bf2059c6d71c00006c,
288
+ 52ec961098d023950500002a, 52e8e98298d0239505000020, 56cae5125795f9a73e000024,
289
+ 530cefaaad0bf1360c000007, 530cefaaad0bf1360c000005, 52d63b2803868f1b0600003a,
290
+ 530cefaaad0bf1360c00000a, 516425ff298dcd4e51000051, 55191149622b194345000010,
291
+ 52fa70142059c6d71c000056, 52f77f4d2059c6d71c00002a, 52efc016c8da89891000001a,
292
+ 52efc001c8da898910000019, 52f896ae2059c6d71c000045, 52eceada98d023950500002d,
293
+ 52efc05cc8da89891000001c, 515e078e298dcd4e51000031, 52fe54252059c6d71c000079,
294
+ 514217a6d24251bc05000005, 52d1389303868f1b06000032, 530cf4d5e2bfff940c000003,
295
+ 52fc946d2059c6d71c000071, 52e8e99e98d0239505000021, 52ef7786c8da898910000015,
296
+ 52d8494698d0239505000007, 530cf51d5610acba0c000001, 52f637972059c6d71c000025,
297
+ 52e9f99798d0239505000025, 515de572298dcd4e51000021, 52fe4ad52059c6d71c000077,
298
+ 52f65bf02059c6d71c000026, 52e8e9d298d0239505000022, 52fa74052059c6d71c00005a,
299
+ 52ffbddf2059c6d71c00007d, 56bc932aac7ad1001900001c, 56c02883ef6e394741000017,
300
+ 52d2b75403868f1b06000035, 52f118aa2059c6d71c000003, 52e929eb98d0239505000023,
301
+ 532c12f2d6d3ac6a3400001d, 52d8466298d0239505000006'
302
+ [3] The 48 questions resulting from merging with their pair have the
303
+ following ids: 5149aafcd24251bc05000045, 515db020298dcd4e51000011,
304
+ 515db54c298dcd4e51000016, 51680a49298dcd4e51000062, 52b06a68f828ad283c000005,
305
+ 52bf1aa503868f1b06000006, 52bf1af803868f1b06000008, 52bf1d6003868f1b0600000e,
306
+ 52cb9b9b03868f1b0600002d, 52d2818403868f1b06000033, 52df887498d023950500000c,
307
+ 52e0c9a298d0239505000010, 52e203bc98d0239505000011, 52e62bae98d0239505000015,
308
+ 52e6c92598d0239505000019, 52e7bbf698d023950500001d, 52ea605098d0239505000028,
309
+ 52ece29f98d023950500002c, 52ecf2dd98d023950500002e, 52ef7754c8da898910000014,
310
+ 52f112bb2059c6d71c000002, 52f65f372059c6d71c000027, 52f77f752059c6d71c00002b,
311
+ 52f77f892059c6d71c00002c, 52f89ee42059c6d71c00004d, 52f89f4f2059c6d71c00004e,
312
+ 52f89fba2059c6d71c00004f, 52f89fc62059c6d71c000050, 52f89fd32059c6d71c000051,
313
+ 52fa6ac72059c6d71c000055, 52fa73c62059c6d71c000058, 52fa73e82059c6d71c000059,
314
+ 52fa74252059c6d71c00005b, 52fc8b772059c6d71c00006e, 52fc94572059c6d71c000070,
315
+ 52fc94ae2059c6d71c000073, 52fc94db2059c6d71c000074, 52fe52702059c6d71c000078,
316
+ 52fe58f82059c6d71c00007a, 530cefaaad0bf1360c000008, 530cefaaad0bf1360c000010,
317
+ 533ba218fd9a95ea0d000007, 534bb147aeec6fbd07000014, 55167dec46478f2f2c00000a,
318
+ 56c04412ef6e39474100001b, 56c1f01eef6e394741000046, 56c81fd15795f9a73e00000c,
319
+ 587d016ed673c3eb14000002
320
+ """
321
+
322
+ _BIOASQ_5B_DESCRIPTION = """\
323
+ The data are intended to be used as training and development data for BioASQ 5,
324
+ which will take place during 2017. There is one file containing the data:
325
+ - BioASQ-trainingDataset5b.json
326
+
327
+ The file contains the data of the first four editions of the challenge: 1799
328
+ questions with their relevant documents, snippets, concepts and rdf triples,
329
+ exact and ideal answers.
330
+ """
331
+
332
+ _BIOASQ_4B_DESCRIPTION = """\
333
+ The data are intended to be used as training and development data for BioASQ 4,
334
+ which will take place during 2016. There is one file containing the data:
335
+ - BioASQ-trainingDataset4b.json
336
+
337
+ The file contains the data of the first three editions of the challenge: 1307
338
+ questions with their relevant documents, snippets, concepts and rdf triples,
339
+ exact and ideal answers from the first two editions and 497 questions with
340
+ similar annotations from the third editions of the challenge.
341
+ """
342
+
343
+ _BIOASQ_3B_DESCRIPTION = """No README provided."""
344
+
345
+ _BIOASQ_2B_DESCRIPTION = """No README provided."""
346
+
347
+ _DESCRIPTION = {
348
+ "bioasq_10b": _BIOASQ_10B_DESCRIPTION,
349
+ "bioasq_9b": _BIOASQ_9B_DESCRIPTION,
350
+ "bioasq_8b": _BIOASQ_8B_DESCRIPTION,
351
+ "bioasq_7b": _BIOASQ_7B_DESCRIPTION,
352
+ "bioasq_6b": _BIOASQ_6B_DESCRIPTION,
353
+ "bioasq_5b": _BIOASQ_5B_DESCRIPTION,
354
+ "bioasq_4b": _BIOASQ_4B_DESCRIPTION,
355
+ "bioasq_3b": _BIOASQ_3B_DESCRIPTION,
356
+ "bioasq_2b": _BIOASQ_2B_DESCRIPTION,
357
+ }
358
+
359
+ _HOMEPAGE = "http://participants-area.bioasq.org/datasets/"
360
+
361
+ # Data access reqires registering with BioASQ.
362
+ # See http://participants-area.bioasq.org/accounts/register/
363
+ _LICENSE = "https://www.nlm.nih.gov/databases/download/terms_and_conditions.html"
364
+
365
+ _URLs = {
366
+ "bioasq_10b": ["BioASQ-training10b.zip", None],
367
+ "bioasq_9b": ["BioASQ-training9b.zip", "Task9BGoldenEnriched.zip"],
368
+ "bioasq_8b": ["BioASQ-training8b.zip", "Task8BGoldenEnriched.zip"],
369
+ "bioasq_7b": ["BioASQ-training7b.zip", "Task7BGoldenEnriched.zip"],
370
+ "bioasq_6b": ["BioASQ-training6b.zip", "Task6BGoldenEnriched.zip"],
371
+ "bioasq_5b": ["BioASQ-training5b.zip", "Task5BGoldenEnriched.zip"],
372
+ "bioasq_4b": ["BioASQ-training4b.zip", "Task4BGoldenEnriched.zip"],
373
+ "bioasq_3b": ["BioASQ-trainingDataset3b.zip", "Task3BGoldenEnriched.zip"],
374
+ "bioasq_2b": ["BioASQ-trainingDataset2b.zip", "Task2BGoldenEnriched.zip"],
375
+ }
376
+
377
+ _SUPPORTED_TASKS = [Tasks.QUESTION_ANSWERING]
378
+ _SOURCE_VERSION = "1.0.0"
379
+ _BIGBIO_VERSION = "1.0.0"
380
+
381
+
382
+ class BioasqTaskBDataset(datasets.GeneratorBasedBuilder):
383
+ """
384
+ BioASQ Task B On Biomedical Semantic QA.
385
+ Creates configs for BioASQ2 through BioASQ10.
386
+ """
387
+
388
+ DEFAULT_CONFIG_NAME = "bioasq_9b_source"
389
+ SOURCE_VERSION = datasets.Version(_SOURCE_VERSION)
390
+ BIGBIO_VERSION = datasets.Version(_BIGBIO_VERSION)
391
+
392
+ # BioASQ2 through BioASQ10
393
+ BUILDER_CONFIGS = []
394
+ for version in range(2, 11):
395
+ BUILDER_CONFIGS.append(
396
+ BigBioConfig(
397
+ name=f"bioasq_{version}b_source",
398
+ version=SOURCE_VERSION,
399
+ description=f"bioasq{version} Task B source schema",
400
+ schema="source",
401
+ subset_id=f"bioasq_{version}b",
402
+ )
403
+ )
404
+
405
+ BUILDER_CONFIGS.append(
406
+ BigBioConfig(
407
+ name=f"bioasq_{version}b_bigbio_qa",
408
+ version=BIGBIO_VERSION,
409
+ description=f"bioasq{version} Task B in simplified BigBio schema",
410
+ schema="bigbio_qa",
411
+ subset_id=f"bioasq_{version}b",
412
+ )
413
+ )
414
+
415
+ def _info(self):
416
+
417
+ # BioASQ Task B source schema
418
+ if self.config.schema == "source":
419
+ features = datasets.Features(
420
+ {
421
+ "id": datasets.Value("string"),
422
+ "type": datasets.Value("string"),
423
+ "body": datasets.Value("string"),
424
+ "documents": datasets.Sequence(datasets.Value("string")),
425
+ "concepts": datasets.Sequence(datasets.Value("string")),
426
+ "ideal_answer": datasets.Sequence(datasets.Value("string")),
427
+ "exact_answer": datasets.Sequence(datasets.Value("string")),
428
+ "triples": [
429
+ {
430
+ "p": datasets.Value("string"),
431
+ "s": datasets.Value("string"),
432
+ "o": datasets.Value("string"),
433
+ }
434
+ ],
435
+ "snippets": [
436
+ {
437
+ "offsetInBeginSection": datasets.Value("int32"),
438
+ "offsetInEndSection": datasets.Value("int32"),
439
+ "text": datasets.Value("string"),
440
+ "beginSection": datasets.Value("string"),
441
+ "endSection": datasets.Value("string"),
442
+ "document": datasets.Value("string"),
443
+ }
444
+ ],
445
+ }
446
+ )
447
+ # simplified schema for QA tasks
448
+ elif self.config.schema == "bigbio_qa":
449
+ features = schemas.qa_features
450
+
451
+ return datasets.DatasetInfo(
452
+ description=_DESCRIPTION[self.config.subset_id],
453
+ features=features,
454
+ supervised_keys=None,
455
+ homepage=_HOMEPAGE,
456
+ license=_LICENSE,
457
+ citation=_CITATION,
458
+ )
459
+
460
+ def _dump_gold_json(self, data_dir):
461
+ """
462
+ BioASQ test data is split into multiple records {9B1_golden.json,...,9B5_golden.json}
463
+ We combine these files into a single test set file 9Bx_golden.json
464
+ """
465
+ version = re.search(r"bioasq_([0-9]+)b", self.config.subset_id).group(1)
466
+ gold_fpath = os.path.join(
467
+ data_dir, f"Task{version}BGoldenEnriched/bx_golden.json"
468
+ )
469
+
470
+ if not os.path.exists(gold_fpath):
471
+ # combine all gold json files
472
+ filelist = glob.glob(os.path.join(data_dir, "*/*.json"))
473
+ data = {"questions": []}
474
+ for fname in sorted(filelist):
475
+ with open(fname, "rt", encoding="utf-8") as file:
476
+ data["questions"].extend(json.load(file)["questions"])
477
+ # dump gold to json
478
+ with open(gold_fpath, "wt", encoding="utf-8") as file:
479
+ json.dump(data, file, indent=2)
480
+
481
+ return f"Task{version}BGoldenEnriched/bx_golden.json"
482
+
483
+ def _split_generators(self, dl_manager):
484
+ """Returns SplitGenerators."""
485
+
486
+ if self.config.data_dir is None:
487
+ raise ValueError(
488
+ "This is a local dataset. Please pass the data_dir kwarg to load_dataset."
489
+ )
490
+
491
+ train_dir, test_dir = dl_manager.download_and_extract(
492
+ [
493
+ os.path.join(self.config.data_dir, _url)
494
+ for _url in _URLs[self.config.subset_id]
495
+ ]
496
+ )
497
+ gold_fpath = self._dump_gold_json(test_dir)
498
+
499
+ # older versions of bioasq have different folder formats
500
+ train_fpaths = {
501
+ "bioasq_2b": "BioASQ_2013_TaskB/BioASQ-trainingDataset2b.json",
502
+ "bioasq_3b": "BioASQ-trainingDataset3b.json",
503
+ "bioasq_4b": "BioASQ-training4b/BioASQ-trainingDataset4b.json",
504
+ "bioasq_5b": "BioASQ-training5b/BioASQ-trainingDataset5b.json",
505
+ "bioasq_6b": "BioASQ-training6b/BioASQ-trainingDataset6b.json",
506
+ "bioasq_7b": "BioASQ-training7b/trainining7b.json",
507
+ "bioasq_8b": "training8b.json", # HACK - this zipfile strips the dirname
508
+ "bioasq_9b": "BioASQ-training9b/training9b.json",
509
+ "bioasq_10b": "BioASQ-training10b/training10b.json",
510
+ }
511
+
512
+ return [
513
+ datasets.SplitGenerator(
514
+ name=datasets.Split.TRAIN,
515
+ gen_kwargs={
516
+ "filepath": os.path.join(
517
+ train_dir, train_fpaths[self.config.subset_id]
518
+ ),
519
+ "split": "train",
520
+ },
521
+ ),
522
+ datasets.SplitGenerator(
523
+ name=datasets.Split.TEST,
524
+ gen_kwargs={
525
+ "filepath": os.path.join(test_dir, gold_fpath),
526
+ "split": "test",
527
+ },
528
+ ),
529
+ ]
530
+
531
+ def _get_exact_answer(self, record):
532
+ """The value exact_answer can be in different formats based on question type."""
533
+ if record["type"] == "yesno":
534
+ exact_answer = [record["exact_answer"]]
535
+ elif record["type"] == "summary":
536
+ exact_answer = []
537
+ # summary question types only have an ideal answer, so use that for bigbio
538
+ if self.config.schema == "bigbio_qa":
539
+ exact_answer = (
540
+ record["ideal_answer"]
541
+ if isinstance(record["ideal_answer"], list)
542
+ else [record["ideal_answer"]]
543
+ )
544
+
545
+ elif record["type"] == "list":
546
+ exact_answer = record["exact_answer"]
547
+ elif record["type"] == "factoid":
548
+ # older version of bioasq sometimes represent this as as string
549
+ exact_answer = (
550
+ record["exact_answer"]
551
+ if isinstance(record["exact_answer"], list)
552
+ else [record["exact_answer"]]
553
+ )
554
+ return exact_answer
555
+
556
+ def _generate_examples(self, filepath, split):
557
+ """Yields examples as (key, example) tuples."""
558
+
559
+ if self.config.schema == "source":
560
+ with open(filepath, encoding="utf-8") as file:
561
+ data = json.load(file)
562
+ for i, record in enumerate(data["questions"]):
563
+ yield i, {
564
+ "id": record["id"],
565
+ "type": record["type"],
566
+ "body": record["body"],
567
+ "documents": record["documents"],
568
+ "concepts": record["concepts"] if "concepts" in record else [],
569
+ "triples": record["triples"] if "triples" in record else [],
570
+ "ideal_answer": record["ideal_answer"]
571
+ if isinstance(record["ideal_answer"], list)
572
+ else [record["ideal_answer"]],
573
+ "exact_answer": self._get_exact_answer(record),
574
+ "snippets": record["snippets"] if "snippets" in record else [],
575
+ }
576
+
577
+ elif self.config.schema == "bigbio_qa":
578
+ with open(filepath, encoding="utf-8") as file:
579
+ uid = 0
580
+ data = json.load(file)
581
+ for record in data["questions"]:
582
+ # for questions that do not have snippets, skip
583
+ if "snippets" not in record:
584
+ continue
585
+ for i, snippet in enumerate(record["snippets"]):
586
+ yield uid, {
587
+ "id": f'{record["id"]}_{i}',
588
+ "document_id": snippet["document"],
589
+ "question_id": record["id"],
590
+ "question": record["body"],
591
+ "type": record["type"],
592
+ "choices": [],
593
+ "context": snippet["text"],
594
+ "answer": self._get_exact_answer(record),
595
+ }
596
+ uid += 1