File size: 3,848 Bytes
48048df 0bcc1f4 ba7558c 48048df ba7558c 48048df ba7558c 48048df 0bcc1f4 | 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 | ---
license:
- mit
- other
license_name: genia-project-license
license_link: LICENSE
features:
- name: input
dtype: string
- name: output
dtype: json
- name: schema
list:
- name: label
dtype: string
- name: description
dtype: string
configs:
- config_name: default
data_files:
- split: train
path: data/genia_ner_train.jsonl
- split: validation
path: data/genia_ner_validation.jsonl
- split: test
path: data/genia_ner_test.jsonl
---
# GENIA NER (mneb format)
The GENIA Term Corpus (Kim et al., 2003) as mneb `entities` NER: 5 biomedical types —
`dna`, `rna`, `protein`, `cell_line`, `cell_type` — over 18,546 PubMed sentences.
| Split | Records | Entities |
|---|---:|---:|
| train | 15,023 | 41,770 |
| validation | 1,669 | 3,947 |
| test | 1,854 | 4,875 |
| **Total** | **18,546** | **50,592** |
- **Format** — one record per GENIA `<sentence>`. Character offsets, end-exclusive
(`input[start:end] == text`). `schema` is always `[]`.
- **Nested entities: outermost only.** GENIA annotates spans that nest —
`"human interferon beta (IFN-beta) regulatory element"` is one DNA span containing two
protein mentions. Only the outer span is kept and inner mentions are dropped, so entities
never overlap. This drops ~12% of protein mentions and ~10% of all mentions, so the release
is **not suitable for nested-NER research** — use the original GENIA XML for that.
- **Sources** — entities from
[`chufangao/GENIA-NER`](https://huggingface.co/datasets/chufangao/GENIA-NER) @
`c2d8d37c233a8abe8635109d106b804db78095d6`, whose BIO tags are already outermost-only and are
used exactly as decoded; `input` text from the original GENIA Term Corpus v3.02 XML, split
positionally (15023/1669/1854) to reproduce chufangao's splits.
- **Replaces a corrupted release.** The previous version fractured every nested entity: its
upstream `extraordinarylab/genia-ner` let an inner entity's `B-` overwrite the outer
entity's, and a naive decoder then started a new span at the orphaned `I-` run. Test row 59's
DNA span came out as `"("` + `") regulatory element"`, with the two inner proteins also
leaking through as their own entities. Rebuilt from scratch and verified: all 18,546 rows
token-aligned, 0 illegal BIO transitions, offset invariant and non-overlap checked on every
entity.
- **License** — annotations MIT (chufangao); text and spans under the GENIA Project License
(PubMed terms; GENIA's own annotations CC-BY 3.0), full text in `LICENSE`. Required
attribution: **"Corpus annotations (c) GENIA Project"**.
Example — test row 59, generated from this release:
```jsonc
{
"input": "The human interferon beta (IFN-beta) regulatory element consists of multiple enhanson domains which are targets for transcription factors involved in inducible expression of the promoter.",
"output": {
"entities": {
"dna": [
{
"text": "human interferon beta (IFN-beta) regulatory element",
"start": 4,
"end": 55
},
{
"text": "enhanson domains",
"start": 77,
"end": 93
},
{
"text": "promoter",
"start": 178,
"end": 186
}
],
"protein": [
{
"text": "transcription factors",
"start": 116,
"end": 137
}
]
}
},
"schema": []
}
```
## Citation
```bibtex
@article{kim2003genia,
title = {GENIA corpus---a semantically annotated corpus for bio-textmining},
author = {Kim, J-D and Ohta, Tomoko and Tateisi, Yuka and Tsujii, Jun'ichi},
journal = {Bioinformatics},
volume = {19},
number = {suppl\_1},
pages = {i180--i182},
year = {2003},
publisher = {Oxford University Press}
}
```
|