During the training process, the model continuously iterated to update the parameters, making the performance continuously optimized; while the Siamese-BiLSTM-Attention-CNN exhibited a large fluctuation in accuracy at the early stage of training, and the loss value showed a decreasing trend as the number of training rounds increased, and the accuracy did not level off until the epoch was 33, with the final result being slightly lower than that of the Siamese-BiGRU-Attention-CNN.
1
—
—