A multi-layer two-dimensional convolutional neural network for sentiment analysis

Abstract

The sentiment analysis method combining long short-term memory (LSTM) network and one-dimensional convolutional neural network (1D-CNN) has become the focus in sentiment analysis tasks. However, 1D-CNN usually captures local features, and LSTM is time-inefficient due to its serialisation characteristics. To this end, we propose a multi-layer two-dimensional convolutional neural network (ML-2D-CNN) for sentiment analysis. In our method, the character-based integer encoding method is used to retain fine-grained sentiment information. Besides, we innovatively introduce interactive features to improve the dimension of feature vectors and enhance sentiment information. The group expansion strategy is used to form several feature mapping groups, which is conducive to the learning of similar sentiment features in the group. Finally, we exploit multi-layer 2D-CNN to extract and mine the fine-grained information of the text. Extensive experiments are conducted on the SST dataset. The experimental results demonstrate that our proposed method is superior to other baseline and state-of-the-art methods.