Table 8 Evaluation results of sentence recognition (SR), medical entity recognition (MER), and keyphrase extraction (KE) using three-way multi-task learning approach. \(^\dagger\) indicates a stochastically dominant performance over STL and Pairwise MTL baselines
From: Sentences, entities, and keyphrases extraction from consumer health forums using multi-task learning
Model | \(\alpha\) | \(\beta\) | SR | MER | KE |
|---|---|---|---|---|---|
\(\text {IndoNLU}_{\text {LARGE}}\) [68] (STL) | \({-}{-}\) | \({-}{-}\) | 92.81 | 59.59 | 46.91 |
Best Pairwise MTL | \({-}{-}\) | \({-}{-}\) | \({\textbf {92.85}}\) | 60.83 | 49.76 |
Parallel Three-way | 0.33 | 0.33 | 92.29 | 60.16 | 48.50 |
0.40 | 0.40 | 92.53 | 60.26 | 45.67 | |
0.40 | 0.20 | 92.41 | 57.73 | \(^\dagger \ {\textbf {51.28}}\) | |
0.20 | 0.40 | 86.70 | 59.10 | 47.32 | |
0.60 | 0.20 | 92.41 | 58.66 | 48.52 | |
0.20 | 0.60 | 86.65 | 60.35 | 43.92 | |
0.20 | 0.20 | 86.59 | 57.91 | 47.34 | |
Hierarchical Three-way | 0.33 | 0.33 | 92.47 | 59.35 | 48.02 |
0.40 | 0.40 | 92.13 | \({\textbf {61.14}}\) | 45.10 | |
0.40 | 0.20 | 92.23 | 58.94 | 50.16 | |
0.20 | 0.40 | 92.44 | 59.87 | 45.64 | |
0.60 | 0.20 | 92.16 | 59.91 | 48.64 | |
0.20 | 0.60 | 89.10 | 54.60 | 42.00 | |
0.20 | 0.20 | 92.17 | 56.89 | 47.25 |