Text Sentiment Analysis Based on CNN-BiLSTM-Attention Model
Fengdong Sun, Na Chu · 2020
To solve the problem that the neural network structure used in the current text sentiment analysis task cannot extract the important features of the text, a model of feature fusion of Convolutional Neural Network(CNN) and Bidirectional Long Short-Term Memory(BiLSTM) is proposed. The CNN is used to extract the local features of the text vector, and the BiLSTM is used to extract the global features related to the text context. The problem that the single CNN model ignores the semantic and grammatical information of words in the context is solved, and the problem of gradient disappearance or gradient diffusion in traditional Recurrent Neural Network(RNN) is effectively avoided. After preprocessing the corpus, the text is expressed as a two-dimensional word vector matrix, and then the local information features are extracted by using the CNN. Then, it is used as the input of BiLSTM to learn the sequence relationship between words and sentences. Then attention mechanism is introduced to highlight important information. The influence on text emotion. Experimental results show that the proposed feature fusion model effectively improves the accuracy of text classification.