Multistream BertGCN for Sentiment Classification Based on Cross-Document Learning

Table 2

Test accuracy of different models.

Model	MR	SST-2
CNN	0.7498	0.8691
LSTM	0.7506	0.8920
Bi-LSTM	0.7768	0.8790
QGRNN	0.7716	0.8403
TextGCN	0.7674	0.8127
SGC	0.7591	—
BERT	0.8570	0.8869
RoBERTa	0.8943	0.8587
BertGCN	0.8600	0.8918
RoBertaGCN	0.8973	0.8833
BertGAT	0.8650	0.8910
RoBertaGAT	0.8921	0.8897
MS-BertGAT	0.8932	0.9090
MS-RoBertaGAT	0.9293	0.8984
MS-BertGCN	0.9042	0.9293
MS-RoBertGCN	0.9316	0.9085

We run all models 10 times and report the mean test accuracy. The bold values given in Table 2 represent the optimal test accuracy for different datasets.