Table 6.

Performance of deep-learning methods with different language representation models on level 1, 2 and 3.

Model	Macro-F1	Macro-Pa	Macro-Rb
BERTc [16]	0.370	0.455	0.381
BERT_IDPd [16]	0.406	0.543 e	0.354
RoBERTaf	0.396	0.503	0.360
RoBERTa_IDP	0.424	0.528	0.386
XLNETg	0.387	0.457	0.336
XLNET_IDP	0.398	0.521	0.364

^aP: precision.

^bR: recall.

^cBERT: bidirectional encoder representations from transformers.

^d_IDP: The model is further trained on the in-domain unlabeled corpus.

^eHighest F1 values are indicated in italics.

^fRoBERTa: robustly optimized bidirectional encoder representations from transformers pretraining approach.

^gXLNET: generalized autoregressive pretraining for language understanding.