Double-Channel Short Text Classification With Fused Enhanced Features

The present paper proposes a double-channel short text classification model with fused enhanced features to tackle the issues of feature sparsity, incomplete feature extraction, and important information loss in short text classification tasks. Specifically, Bert and CNN are employed to extract the global and local features of short text, respectively. Moreover, an attention mechanism is incorporated to enhance channel features and mitigate the problem of feature sparsity in short text. To prevent the loss of important information, the SoftPool pooling mechanism is adopted instead of the traditional average pooling. Then, the global and local features are fused interactively to address the issue of incomplete feature extraction. Finally, the Softmax classifier is utilized to obtain the final results for short text classification. The experimental results indicate that the model proposed in this study achieved optimal classification performance across three short-text datasets.