From: AraXLNet: pre-trained language model for sentiment analysis of Arabic
Dataset
BERT (%)
XLNet (%)
SQiAD1.1
92.8
94.0
SQuAD2.0
85.5
87.8
RACE
75.1
77.4
MNLI
87.3
88.4
QNLI
93.0
93.9
QQP
91.4
91.8
RTE
74.0
94.4