Add dataset card
Browse files
README.md
CHANGED
|
@@ -1,4 +1,8 @@
|
|
| 1 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
| 2 |
dataset_info:
|
| 3 |
- config_name: crossner_ai
|
| 4 |
features:
|
|
@@ -122,3 +126,28 @@ configs:
|
|
| 122 |
- split: test
|
| 123 |
path: crossner_science/test-*
|
| 124 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
language: [en]
|
| 4 |
+
tags: [ner, crossner, evaluation]
|
| 5 |
+
pretty_name: UniversalMedNER – CrossNER evaluation subsets
|
| 6 |
dataset_info:
|
| 7 |
- config_name: crossner_ai
|
| 8 |
features:
|
|
|
|
| 126 |
- split: test
|
| 127 |
path: crossner_science/test-*
|
| 128 |
---
|
| 129 |
+
|
| 130 |
+
# UniversalMedNER – CrossNER evaluation subsets
|
| 131 |
+
|
| 132 |
+
The five CrossNER target domains ([DFKI-SLT/cross_ner](https://huggingface.co/datasets/DFKI-SLT/cross_ner),
|
| 133 |
+
Liu et al., 2021) in exactly the format of `universalmedner/bioner-eval-subset`, so the same
|
| 134 |
+
evaluation code scores them: same system and user prompt, same construction of the requested types
|
| 135 |
+
(up to 6: a random number of the row's present types plus absent distractors drawn with inverse
|
| 136 |
+
frequency from the domain's train split), plus `raw_text` and `all_gold`.
|
| 137 |
+
|
| 138 |
+
| Config | Rows (full test set) |
|
| 139 |
+
|---|---|
|
| 140 |
+
| `crossner_ai` | 431 |
|
| 141 |
+
| `crossner_literature` | 416 |
|
| 142 |
+
| `crossner_music` | 465 |
|
| 143 |
+
| `crossner_politics` | 651 |
|
| 144 |
+
| `crossner_science` | 543 |
|
| 145 |
+
|
| 146 |
+
- Labels converted to the evaluation's UPPER_SNAKE style (`programlang` → `PROGRAMMING_LANGUAGE`,
|
| 147 |
+
`musicalartist` → `MUSICAL_ARTIST`, ...).
|
| 148 |
+
- `misc` is never requested and is not in the prompt gold (as in UniversalNER's benchmark); its spans
|
| 149 |
+
remain in `all_gold` under `MISC`.
|
| 150 |
+
- The system prompt is the benchmarks' one verbatim (it mentions biomedical text), so the dataset is the
|
| 151 |
+
only thing that changes with respect to the biomedical evaluation.
|
| 152 |
+
|
| 153 |
+
Built with `04_evaluation/build_crossner_eval.py` (branch `medgemma-recovery`).
|