Interactive Dual Attention Network for Text Sentiment Classification

Table 3

Performance comparison with baseline methods.

Approach	ChnSentiCorp		NLPCC-CN		NLPCC-EN		MR
	Accuracy	Macro-F1	Accuracy	Macro-F1	Accuracy	Macro-F1	Accuracy	Macro-F1
SVM	0.8618	0.8528	0.7479	0.7441	0.8226	0.8143	0.7914	0.7852
LSTM	0.8681	0.8570	0.7572	0.7557	0.8381	0.8379	0.7844	0.7705
BiLSTM	0.8831	0.8693	0.7603	0.7573	0.8488	0.8477	0.7941	0.7877
ATT-BiLSTM	0.8945	0.8892	0.7665	0.7585	0.8503	0.8491	0.7952	0.7909
H-RNN-CNN	0.8940	0.9030	0.7550	0.7790	—	—	0.8190	—
CRNN	0.9108	0.9082	0.7702	0.7648	0.8579	0.8456	0.8228	—
fastText	0.9203	0.9170	0.7706	0.7624	0.8670	0.8615	0.8181	0.8121
LR-BiLSTM	—	—	—	—	—	—	0.8210	—
IDAN	0.9297	0.9293	0.8005	0.7875	0.9181	0.9068	0.8266	0.8135

Bold values indicate the best performances.