Update the card: consistent title, project and phase, metric, related resources
Browse files
README.md
CHANGED
|
@@ -1,19 +1,33 @@
|
|
| 1 |
---
|
| 2 |
-
tags: [named-entity-recognition]
|
|
|
|
|
|
|
| 3 |
---
|
| 4 |
-
|
| 5 |
-
|
| 6 |
-
|
| 7 |
-
|
| 8 |
-
|
| 9 |
-
|
| 10 |
-
`
|
| 11 |
-
|
| 12 |
-
|
| 13 |
-
|
| 14 |
-
|
| 15 |
-
|
| 16 |
-
|
| 17 |
-
|
| 18 |
-
|
| 19 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
+
tags: [named-entity-recognition, evaluation, chain-of-thought]
|
| 3 |
+
language: [en]
|
| 4 |
+
pretty_name: UniversalMedNER – Evaluation Sets of the Free vs Structured Reasoning Experiment
|
| 5 |
---
|
| 6 |
+
|
| 7 |
+
# UniversalMedNER — Evaluation Sets of the Free vs Structured Reasoning Experiment
|
| 8 |
+
|
| 9 |
+
Part of **UniversalMedNER**, the code, data and models of the bachelor thesis *Structuring Universal Chain-of-thought for Named Entity Recognition* (Department of Computer Science and Engineering, University of Bologna, 2026).
|
| 10 |
+
|
| 11 |
+
The query sets on which the second-phase students (`universalmedner/qwen3-4b-thinking-free-cot-lora`,
|
| 12 |
+
`universalmedner/qwen3-4b-thinking-structured-cot-lora`) and the base model are evaluated.
|
| 13 |
+
|
| 14 |
+
## Query protocol
|
| 15 |
+
|
| 16 |
+
For every sentence: all the types of its gold annotation, plus 1-3 **distractors** (number drawn uniformly) among the
|
| 17 |
+
labels of the same domain or dataset that do not occur in it; the order of the types is shuffled; one fixed seed (42),
|
| 18 |
+
so that every model answers exactly the same queries.
|
| 19 |
+
|
| 20 |
+
## Files
|
| 21 |
+
|
| 22 |
+
- `crossner/`: the five CrossNER domains (AI, literature, music, politics, science), full test sets
|
| 23 |
+
(`DFKI-SLT/cross_ner`); labels in upper case with underscores, `MISC` included; text and gold mentions
|
| 24 |
+
detokenized with Moses, so that every gold mention is a substring of the text.
|
| 25 |
+
- `biomedical/`: the nine biomedical benchmarks, with the same 500 sentences per dataset as
|
| 26 |
+
`universalmedner/bioner-eval-subset` (465 for JNLPBA-Rare). Distractors are drawn only from the labels of the same
|
| 27 |
+
dataset, because a label of another dataset is not annotated there.
|
| 28 |
+
|
| 29 |
+
Each row: `id`, `text`, `requested_types`, `distractors`, `gold` ({type: [mentions]}), `free_prompt`,
|
| 30 |
+
`structured_prompt` (the exact prompt templates of `disi-unibo-nlp-students/pilener-teacher-cot-agreement-v1`).
|
| 31 |
+
The in-domain test set is the test split of that dataset and is not duplicated here.
|
| 32 |
+
|
| 33 |
+
Built with `04_evaluation/build_tutor_eval.py` and `04_evaluation/build_tutor_eval_bio.py`.
|