Table 9.
Holdout dataset metrics of all models tested on different datasets
| Dataset and model used for training | Precision | Recall | F1-score |
|---|---|---|---|
| Co-occurrence (bag level) | 0.48 | 1.00 | 0.65 |
| FULL | |||
| BAG_BIOBERT | 0.62 | 0.55 | 0.58 |
| BAG_PCNN | 0.65 | 0.56 | 0.60 |
| BIOBERT | 0.59 | 0.95 | 0.73 |
| PCNN | 0.55 | 0.98 | 0.70 |
| MULTI | |||
| BAG_BIOBERT | 0.61 | 0.62 | 0.62 |
| BAG_PCNN | 0.62 | 0.63 | 0.63 |
| BIOBERT | 0.60 | 0.95 | 0.74 |
| PCNN | 0.56 | 0.97 | 0.71 |
| MULTI_LIMITED | |||
| BAG_BIOBERT | 0.65 | 0.66 | 0.66 |
| BAG_PCNN | 0.61 | 0.66 | 0.63 |
| BIOBERT | 0.56 | 0.94 | 0.70 |
| PCNN | 0.52 | 0.99 | 0.68 |
| SINGLE | |||
| BAG_BIOBERT | 0.55 | 0.88 | 0.68 |
| BAG_PCNN | 0.48 | 1.00 | 0.65 |
| BIOBERT | 0.48 | 1.00 | 0.65 |
| PCNN | 0.48 | 1.00 | 0.65 |