BERT-CNN-BiLSTM: A Hybrid Deep Learning Model for Accurate Sentiment Analysis

Comments can influence people's choices. It is important to accurately determine the sentiment polarity of comments. Around this problem, this paper proposed a sentiment analysis method based on BERT-CNN-BiLSTM model. It first used Bidirectional Encoder Representations from Transformers (BERT) to transform the words in the input sequence into a vector representation. Then, Convolutional Neural Network (CNN) was used to extract features, i.e., the output vector of BERT was convolved along the dimension of sequence length to further extract the features in the sequence. Next, Bidirectional Long Short-Term Memory (BiLSTM) was used to further encode the features and capture the long-term dependencies in the sequence. Finally, the output vector of the BiLSTM was fed into a fully connected layer to make classification predictions. Accuracy was used as an evaluation metric to monitor the performance of the model. Compared with BERT, BiLSTM, CNN, and BERT-BiLSTM models, the accuracy of this model has been improved by 6.98%, 21.05%, 27.78%, and 3.37%, respectively.