merionum commited on
Commit
ea955a1
·
verified ·
1 Parent(s): 11afe11

Add RuClinNER: three Russian clinical NER corpora (parquet + card)

Browse files
README.md ADDED
@@ -0,0 +1,205 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: cc-by-nc-sa-4.0
3
+ language:
4
+ - ru
5
+ multilinguality:
6
+ - monolingual
7
+ annotations_creators:
8
+ - expert-generated
9
+ source_datasets:
10
+ - original
11
+ task_categories:
12
+ - token-classification
13
+ task_ids:
14
+ - named-entity-recognition
15
+ pretty_name: RuClinNER
16
+ size_categories:
17
+ - 1K<n<10K
18
+ tags:
19
+ - clinical
20
+ - medical
21
+ - biomedical
22
+ - named-entity-recognition
23
+ - nested-ner
24
+ - relation-extraction
25
+ - entity-linking
26
+ - icd-10
27
+ - russian
28
+ configs:
29
+ - config_name: medterm
30
+ data_files:
31
+ - split: train
32
+ path: data/medterm/train.parquet
33
+ - split: test
34
+ path: data/medterm/test.parquet
35
+ - config_name: drugrel
36
+ data_files:
37
+ - split: train
38
+ path: data/drugrel/train.parquet
39
+ - split: test
40
+ path: data/drugrel/test.parquet
41
+ - config_name: covid_rmj
42
+ data_files:
43
+ - split: train
44
+ path: data/covid_rmj/train.parquet
45
+ - split: test
46
+ path: data/covid_rmj/test.parquet
47
+ ---
48
+
49
+ # RuClinNER
50
+
51
+ RuClinNER is a collection of three complementary Russian **clinical** named-entity-recognition
52
+ corpora. They span distinct kinds of clinical text written in healthcare settings —
53
+ patient-reported complaints, physician prescriptions, and peer-reviewed clinical research
54
+ articles — and together contain **55,185 entity spans across 22 entity types** and
55
+ **18,112 relations across 8 relation types** over **6,797 records**.
56
+
57
+ The collection accompanies the paper *RuClinNER: Russian Clinical Multi-Label NER Annotated
58
+ Corpora*. It is intended for research on Russian clinical and biomedical NLP, where annotated
59
+ clinical text remains scarce: existing Russian resources largely target consumer drug reviews
60
+ or scientific abstracts rather than text produced in clinical practice.
61
+
62
+ ## The three corpora
63
+
64
+ | Config | Text type | Records | Entities | Types | Relations | Median words | Notable |
65
+ |---|---|--:|--:|--:|--:|--:|---|
66
+ | `medterm` | patient complaints | 4,474 | 12,582 | 3 | — | 9 | ICD-10 normalization subset |
67
+ | `drugrel` | drug prescriptions | 2,200 | 35,635 | 13 | 18,112 | 28 | nested entities + relations; patient metadata |
68
+ | `covid_rmj` | clinical journal articles | 123 | 6,968 | 6 | — | 2,386 | long documents; dual annotation layers |
69
+ | **Total** | | **6,797** | **55,185** | **22** | **18,112** | | |
70
+
71
+ - **MedTerm** — short, free-text symptom narratives from a symptom-checking service.
72
+ Three entity types (`SYMPTOM`, `RISK`, `DISEASE`). A curated subset of entities is linked
73
+ to **ICD-10** canonical concepts (Russian and English names), supporting clinical concept
74
+ normalization / entity linking.
75
+ - **DrugRel** — Russian clinical prescription records. Thirteen entity types covering drug
76
+ names, dosage components, and modifiers, with **hierarchically nested** entities: a `REGIME`
77
+ span is a *container* whose tokens also carry nested labels (`DOSE`, `DOSIR`, `FREQ`, `DUR`,
78
+ `TIME`, `USE`, …). Eight relation types link attributes to their drug. Records are linked to
79
+ de-identified patient metadata (age, gender, ICD-10 diagnosis codes).
80
+ - **COVID-RMJ** — full-length COVID-19 articles from the Russian Medical Journal, annotated by
81
+ medical experts over six entity types. Two parallel annotation layers are provided: the full
82
+ article text and a structured abstract, each with character spans and pre-computed BIO tags.
83
+
84
+ ## Supported tasks
85
+
86
+ The corpora support several clinical-NLP tasks:
87
+
88
+ - **Named entity recognition** — all three corpora (character-offset spans).
89
+ - **Nested / multi-label token classification** — DrugRel `REGIME` containers (≈22,900
90
+ overlapping span pairs); a single token may carry more than one label.
91
+ - **Relation extraction** — DrugRel drug–attribute relations (8 types).
92
+ - **Clinical concept normalization / entity linking** — MedTerm entities linked to ICD-10.
93
+ - **Long-document NER** — COVID-RMJ articles (median ≈2,400 words).
94
+ - **Cross-corpus and domain-transfer studies**, and **evaluation of LLM prompting** for
95
+ clinical information extraction across complementary text types.
96
+
97
+ ## Usage
98
+
99
+ ```python
100
+ from datasets import load_dataset
101
+
102
+ medterm = load_dataset("medlinx/RuClinNER", "medterm")
103
+ drugrel = load_dataset("medlinx/RuClinNER", "drugrel")
104
+ covid = load_dataset("medlinx/RuClinNER", "covid_rmj")
105
+
106
+ ex = medterm["test"][0]
107
+ print(ex["text"])
108
+ for e in ex["entities"]:
109
+ print(e["label"], ":", e["text"], e.get("coding"))
110
+ ```
111
+
112
+ ## Data fields
113
+
114
+ **`medterm`**
115
+ - `id` (string), `text` (string)
116
+ - `entities`: list of `{start, end, label, text, coding}` where `coding` is
117
+ `{icd10, name_ru, name_en}` for the normalized subset, else `null`.
118
+
119
+ **`drugrel`**
120
+ - `id`, `record_id` (string), `text` (string)
121
+ - `entities`: list of `{id, start, end, label, text}`
122
+ - `relations`: list of `{type, from_id, to_id, from_text, to_text}`
123
+ - `patient_age` (int, nullable), `patient_gender` (string, nullable)
124
+ - `diagnosis` (string), `diagnosis_icd_codes` (list of string)
125
+
126
+ **`covid_rmj`**
127
+ - `id`, `text` (string)
128
+ - `text_entities`: list of `{start, end, label, text}` over `text`
129
+ - `text_tokens`, `text_ner_tags`: pre-tokenized text with BIO tags
130
+ - `annotation` (string, structured abstract) with `annotation_entities`,
131
+ `annotation_tokens`, `annotation_ner_tags`
132
+
133
+ ## Splits
134
+
135
+ | Config | train | test |
136
+ |---|--:|--:|
137
+ | `medterm` | 3,049 | 1,425 |
138
+ | `drugrel` | 2,100 | 100 |
139
+ | `covid_rmj` | 98 | 25 |
140
+
141
+ For MedTerm, the ICD-10 normalization subset (508 records, 1,294 linked entities) falls within
142
+ the `test` split. For COVID-RMJ, a train/test split is provided here; because the corpus is
143
+ small (123 documents), the accompanying paper instead evaluates encoders with 5-fold
144
+ cross-validation.
145
+
146
+ ## Entity and relation types
147
+
148
+ **MedTerm** — `SYMPTOM` (symptoms/signs), `RISK` (risk factors), `DISEASE` (named conditions).
149
+
150
+ **DrugRel** — `DRUG`, `REGIME` (full dosage regimen; container), `DOSE` (numeric dose),
151
+ `DOSIR` (per-intake instruction), `FREQ`, `DUR`, `TIME`, `USE` (route), `FORM`, `COND`
152
+ (condition for taking), `INN` (active substance), `NUM`, `NEG`. Relations:
153
+ `REGIME-DRUG`, `FORM-DRUG`, `DOSE-DRUG`, `USE-DRUG`, `COND-DRUG`, `INN-DRUG`, `NEG-DRUG`, `ALT`.
154
+
155
+ **COVID-RMJ** — `DISEASE`, `PROCEDURE`, `SYMPTOM`, `DRUG`, `ANATOMY`, `LAB_VALUE`.
156
+
157
+ ## Dataset creation
158
+
159
+ Each corpus was annotated by domain experts (medical informaticians, clinical pharmacologists,
160
+ and medical specialists) following written guidelines. Records were independently annotated and
161
+ disagreements adjudicated. Inter-annotator agreement (Cohen's κ): MedTerm 0.87 (boundaries) /
162
+ 0.91 (labels); DrugRel 0.89 (entities) / 0.81 (relations). COVID-RMJ was reviewed by at least
163
+ two experts with adjudication and passes automated boundary/overlap/BIO consistency checks.
164
+
165
+ ## Considerations for using the data
166
+
167
+ - **Class imbalance** is substantial and corpus-specific (e.g. `SYMPTOM` ≈85% of MedTerm
168
+ spans; `DISEASE` ≈61% of COVID-RMJ spans). Report macro- as well as micro-averaged metrics.
169
+ - **MedTerm** contains genuine negatives: ≈18% of records have no annotated entity.
170
+ - **DrugRel** includes 67 relations with `type=null` (annotation artifacts); exclude these for
171
+ relation-extraction evaluation. The `RISK`/`DISEASE` distinction can be semantically
172
+ ambiguous (e.g. obesity).
173
+ - **DrugRel demographics** skew toward female patients (≈74%), with many pediatric and
174
+ obstetric records; consider this when transferring models to other populations.
175
+ - **COVID-RMJ** entity density varies ~100× across documents.
176
+
177
+ ## Personal and sensitive information
178
+
179
+ The data is de-identified: MedTerm and DrugRel records contain no patient names, dates of
180
+ birth, or contact information; COVID-RMJ articles are published, peer-reviewed texts with no
181
+ patient-level data. DrugRel patient metadata is limited to coarse age, gender, and ICD-10
182
+ diagnosis codes.
183
+
184
+ ## Licensing
185
+
186
+ Released under **Creative Commons Attribution-NonCommercial-ShareAlike 4.0 (CC BY-NC-SA 4.0)**
187
+ for research use.
188
+
189
+ ## Citation
190
+
191
+ ```bibtex
192
+ @misc{ruclinner2026,
193
+ title = {RuClinNER: Russian Clinical Multi-Label NER Annotated Corpora},
194
+ author = {Gudkov, Vadim and Varennikova, Anastasia and Miroshnichenko, Polina and
195
+ Mitrofanova, Olga and Gousyatskaya, Polina and Boitsova, Daria},
196
+ year = {2026},
197
+ publisher = {Hugging Face},
198
+ howpublished = {\url{https://huggingface.co/datasets/medlinx/RuClinNER}}
199
+ }
200
+ ```
201
+
202
+ ## Contact
203
+
204
+ Maintained by [Medlinx](https://huggingface.co/medlinx). Questions: `gudkov.v@medlinx.online`.
205
+ </content>
data/covid_rmj/test.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:312b6b20b249d94e19aa76bb941ee5f283cfb8fa2b0930e7363eb58e1df541a9
3
+ size 2270496
data/covid_rmj/train.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5cca7f2a0ef5f1fdb67a660ba30b28275e6a1cc2dca9ff988c28aeb5c41f3436
3
+ size 8560325
data/drugrel/test.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4fa5c8fca3a490bf00685d6164bdb6af79916343beb6d29e89d15244dc5966af
3
+ size 214397
data/drugrel/train.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4392dec52ac7e4305815173a6e167962f55b216848b25f7eec13f16a9606043a
3
+ size 4013866
data/medterm/test.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2d2dc278339640a9dffbe03f69c8b8206faae3f816341eea7afa074f62cd29a3
3
+ size 349545
data/medterm/train.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e528f242914edfcfce710397299fec9e009b17c5ff122d1828fa7709ff374a32
3
+ size 761841