Table 6 Evaluation results of sentence recognition (SR), medical entity recognition (MER), and keyphrase extraction (KE) using single-task learning approach. \(^\dagger\) indicates a stochastically dominant performance over other models
From: Sentences, entities, and keyphrases extraction from consumer health forums using multi-task learning
Model/Encoder | SR | MER | KE |
|---|---|---|---|
CRFs | 64.61 | 36.75 | 19.87 |
IndoDistilBERT1 | 92.35 | 54.43 | 42.64 |
\(\text {IndoLEM}_{\text {BASE}}\) [67] | 93.26 | 56.93 | \(^\dagger \ {\textbf {47.48}}\) |
\(\text {IndoNLU}_{\text {BASE}}\) [68] | 92.08 | 54.58 | 43.46 |
\(\text {IndoNLU}_{\text {BASE}}\) FT2 | 92.14 | 54.07 | 43.44 |
IndoBERTweet [69] | 92.77 | 53.25 | 42.63 |
\(\text {IndoNLU}_{\text {LARGE}}\) [68] | 92.81 | \({\textbf {59.59}}\) | 46.91 |
XLM-MLM [70] | 90.21 | 54.81 | 38.40 |
\(\text {XLM-R}_{\text {BASE}}\) [71] | \(^\dagger \ {\textbf {93.70}}\) | 45.78 | 46.55 |
\(\text {XLM-R}_{\text {LARGE}}\) [71] | 93.37 | 59.32 | 43.03 |