Toward multi-label sentiment analysis: a transfer learning based approach

Journal of Big Data

Table 10 Experiment results of wine reviews

Model	Accuracy (%)	Hamming loss	Macro F1	Micro F1
Proposed models
BERT	79.13	0.021	0.86	0.92
XLNet	78.41	0.021	0.86	0.92
Baseline deep learning models
LSTM	58.01	0.037	0.72	0.85
BiLSTM	56.75	0.039	0.73	0.85
CNN + LSTM	51.92	0.042	0.64	0.83
Baseline machine learning models
SGD + OR	34.10	0.053	0.58	0.77
LR + OR	38.03	0.049	0.65	0.80
SVC + OR	47.92	0.041	0.77	0.84
RF + OR	64.97	0.029	0.83	0.88
SGD + BR	33.81	0.053	0.58	0.77
LR + BR	38.03	0.049	0.64	0.80
SVC + BR	47.92	0.041	0.77	0.84
RF + BR	63.35	0.030	0.82	0.88
SGD + CC	50.51	0.056	0.62	0.78
LR + CC	54.24	0.052	0.67	0.80
SVC + CC	64.12	0.039	0.79	0.85
RF + CC	68.06	0.030	0.83	0.88
SGD + LP	58.25	0.051	0.72	0.80
LR + LP	58.16	0.049	0.84	0.87
SVC + LP	70.94	0.034	0.84	0.87
RF + LP	72.54	0.035	0.82	0.87
SGD + RakelD	46.95	0.052	0.70	0.80
LR + RakelD	56.13	0.049	0.68	0.81
SVC + RakelD	57.47	0.038	0.83	0.85
RF + RakelD	70.82	0.034	0.83	0.87
BRkNNa	45.91	0.062	0.67	0.76
BRkNNb	46.13	0.060	0.66	0.77
MLARAM	50.53	0.044	0.71	0.79
MLkNN	48.25	0.055	0.68	0.77