filippogreppi commited on
Commit
c191aca
·
verified ·
1 Parent(s): b31923a

Update the card: consistent title, project and phase, metric, related resources

Browse files
Files changed (1) hide show
  1. README.md +31 -17
README.md CHANGED
@@ -1,19 +1,33 @@
1
  ---
2
- tags: [named-entity-recognition]
 
 
3
  ---
4
- # Evaluation sets of the free vs structured CoT experiment
5
-
6
- Query protocol asked by the tutors: for every sentence, all the types of its gold annotation plus 1-3
7
- distractors (number drawn uniformly) among the labels of the same domain/dataset absent from the
8
- sentence, shuffled, fixed seed (42). Each row: `id`, `text`, `requested_types`, `distractors`,
9
- `gold` ({type: [mentions]}), `free_prompt`, `structured_prompt` (the exact templates of
10
- `pilener-teacher-cot-agreement-v1`).
11
-
12
- - `crossner/`: the 5 CrossNER domains (ai, literature, music, politics, science), full test splits
13
- (DFKI-SLT/cross_ner), labels in UPPER_SNAKE, MISC included; text and gold mentions detokenized with Moses.
14
- - `biomedical/`: the 9 BioNER datasets, the same 500 test sentences per dataset as
15
- `universalmedner/bioner-eval-subset` (465 for JNLPBA-Rare); distractors only from the same dataset's
16
- labels (a label of another dataset is not annotated there).
17
- - The Pile-NER test split is not included: it is the test split of the tutors' dataset.
18
-
19
- Built by `04_evaluation/build_tutor_eval.py` and `build_tutor_eval_bio.py`.
 
 
 
 
 
 
 
 
 
 
 
 
 
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`.