Dual-channel Text Classification Based on the Attention Mechanism

Text classification can realize information organization and induction, information retrieval and search, sentiment analysis, spam filtering, automated text processing, intelligence collection and security. As the text information to be processed becomes increasingly complex, using traditional convolutional neural networks and recurrent neural networks alone is far from satisfactory for addressing the allocation of text information weights. Based on this, this paper improves the text classification model by replacing the convolutional neural network with a bidirectional convolutional neural network as one channel, replacing the recurrent neural network with a bidirectional long short-term memory network as another channel, and injecting attention mechanisms into the two channels respectively. Bi-LSTM channels can more effectively extract contextual information from news texts, while Bi-CNN channels can more effectively extract local feature information from news texts. After both channels undergo individual attention mechanism processing, the features are fused. The fused result is then passed to the softmax function to accomplish the final news text classification task. This paper conducts performance evaluation on two publicly available datasets, namely Today's Headlines and THUCNews. Experimental results demonstrate that the proposed Bi-Modal Text Fusion Network (BiCTFN) outperforms other classification models and exhibits superior classification performance for long text classification tasks.