wzkariampuzha commited on
Commit
40b0209
1 Parent(s): fff4533

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +3 -3
README.md CHANGED
@@ -12,7 +12,7 @@ licenses:
12
  multilinguality:
13
  - monolingual
14
  size_categories:
15
- - 10K<n<100K **FIX**
16
  task_categories:
17
  - structure-prediction
18
  task_ids:
@@ -84,7 +84,7 @@ Comparing the programmatically labeled test set to the manually corrected test s
84
  |:----------------:|:------------------------:|:---------:|:------:|:-----:|
85
  | Entity-Level | Overall | 0.559 | 0.662 | 0.606 |
86
  | | Location | 0.597 | 0.661 | 0.627 |
87
- | | Epidemiologic Identifier | 0.854 | 0.911 | 0.882 |
88
  | | Epidemiologic Rate | 0.175 | 0.255 | 0.207 |
89
  | Token-Level | Overall | 0.805 | 0.710 | 0.755 |
90
  | | Location | 0.868 | 0.713 | 0.783 |
@@ -134,7 +134,7 @@ Assisting 25-30 millions Americans with rare diseases. Additionally can be usefu
134
  - 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.
135
  - Tokenization was done by spaCy which may be a limitation (or not) for current and future models trained on this set.
136
  - 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.
137
- - 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.
138
 
139
  ## Additional Information
140
 
 
12
  multilinguality:
13
  - monolingual
14
  size_categories:
15
+ - 100K<n<1M
16
  task_categories:
17
  - structure-prediction
18
  task_ids:
 
84
  |:----------------:|:------------------------:|:---------:|:------:|:-----:|
85
  | Entity-Level | Overall | 0.559 | 0.662 | 0.606 |
86
  | | Location | 0.597 | 0.661 | 0.627 |
87
+ | | Epidemiologic Type | 0.854 | 0.911 | 0.882 |
88
  | | Epidemiologic Rate | 0.175 | 0.255 | 0.207 |
89
  | Token-Level | Overall | 0.805 | 0.710 | 0.755 |
90
  | | Location | 0.868 | 0.713 | 0.783 |
 
134
  - 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.
135
  - Tokenization was done by spaCy which may be a limitation (or not) for current and future models trained on this set.
136
  - 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.
137
+ - 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.
138
 
139
  ## Additional Information
140