Table 2 Parameter settings for all models
Baseline model | Parameter setting | |
|---|---|---|
SVM | C = 5.0, gamma = 0.01, kernel:rbf | |
Random Forest | N_estimators = 500, class_weight = balanced, max_depth = 20 | |
Naive Bayes (Gaussian) | Default | |
Naive Bayes (multinominal) | Default | |
Proposed model | Layers | Parameter setting |
|---|---|---|
Self-taught CNN (st-CNN) | Embedding | Size: 300, max_length: 20 |
Dropout | Dropout_rate: 0.2 | |
Convolutional | Kernel_sizes: [2,3,4], number_kernels: 20 activation_function: Relu, strides: 1 | |
Max pooling | Pool_size: 2 | |
Flatten | No parameter | |
Concatenate | No parameter | |
Dropout | Dropout_rate: 0.5 | |
Two dense layers | Dense_layer_1: size: 520 × 500; dense_layer_2: size: 500 × 2 | |
Self-taught LSTM (st-LSTM) | Embedding | Size: 300, max_length: 20 |
Dropout | Dropout_rate: 0.2 | |
LSTM | Sequence_output: false | |
Dropout | Dropout_rate: 0.5 | |
Two dense layers | Dense_layer_1: size: 300 × 500; dense_layer_2: size: 500 × 2 |