Sentiment analysis has been a hot research topic in NLP and data mining fields in the last decade. To solve the feature dimension is too high and the pool layer information is lost, which leads to the loss of the details of the emotion vocabulary. I proposed a Word2vec-CNN-BiLSTM hybrid model means the Word Vector Model, Bidirectional Long-term and Short-term Memory networks and convolutional neural network are combined in Quora dataset. The experiment shows that the accuracy achieved 91.48% performs better than each single model in short text. However, even with the hybrid approach that leverages the powers of these two deep-learning models, the number of features remains huge and hindering the training process. Secondly, I propose an attention-based CNN-BiLSTM hybrid model in IMDB movie reviews dataset. When the data size was 13 k, the proposed model had the highest accuracy at 0.908, and the F1 score also showed the highest performance at 0.883. When data was 20 k, the F1 score showed the best performance at 0.906, and accuracy was the highest at 0.929. The experimental results show that the BiLSTM-CNN model based on attention mechanism can effectively improve the performance of sentiment classification when processing long-text tasks.
Yue WANG (Thu,) studied this question.