Update README.md

README.md

@@ -12,7 +12,7 @@ licenses:
 multilinguality:
 - monolingual
 size_categories:
-- 10K<n<100K **FIX**
+- 100K<n<1M
 task_categories:
 - structure-prediction
 task_ids:
@@ -84,7 +84,7 @@ Comparing the programmatically labeled test set to the manually corrected test s
 |:----------------:|:------------------------:|:---------:|:------:|:-----:|
 |   Entity-Level   |          Overall         |   0.559   |  0.662 | 0.606 |
 |                  |         Location         |   0.597   |  0.661 | 0.627 |
-|                  | Epidemiologic Identifier |   0.854   |  0.911 | 0.882 |
+|                  | Epidemiologic Type       |   0.854   |  0.911 | 0.882 |
 |                  |    Epidemiologic Rate    |   0.175   |  0.255 | 0.207 |
 |   Token-Level    |          Overall         |   0.805   |  0.710 | 0.755 |
 |                  |         Location         |   0.868   |  0.713 | 0.783 |
@@ -134,7 +134,7 @@ Assisting 25-30 millions Americans with rare diseases. Additionally can be usefu
 - The [long short-term memory recurrent neural network epi classifier](https://pubmed.ncbi.nlm.nih.gov/34457147/) was used to sift the 7699 rare disease abstracts. This model had a hold-out validation F1 score of 0.886 and a test F1 (which was compared against a GARD curator who used full-text articles to determine truth-value of epidemiological abstract) of 0.701. With 620 epi abstracts filtered from 7699 original rare disease abstracts, there are likely several false positives and false negative epi abstracts.
 - Tokenization was done by spaCy which may be a limitation (or not) for current and future models trained on this set. 
 - The programmatic labeling was very imprecise as seen by Table 1. This is likely the largest limitation of the [BioBERT-based model](https://huggingface.co/ncats/EpiExtract4GARD) trained on this set. 
-- The test set was difficult to validate even for general NCATS researchers, which is why we relied on a rare disease expert to verify our modifications. As this task of epidemiological information identification is quite difficult for non-expert humans to complete, this set, and especially a gold-standard dataset in the possible future, represents a challenging gauntlet for NLP systems to compete on. 
+- The test set was difficult to validate even for general NCATS researchers, which is why we relied on a rare disease expert to verify our modifications. As this task of epidemiological information identification is quite difficult for non-expert humans to complete, this set, and especially a gold-standard dataset in the possible future, represents a challenging gauntlet for NLP systems, especially those focusing on numeracy, to compete on. 
 
 ## Additional Information