Table 2 Main neural network architectures in the HFC-AES model.
Model architecture | Core function | Advantage | Application locations in this study |
|---|---|---|---|
Convolutional Neural Network (CNN) | Extract local semantic features of text | It is good at capturing local patterns, with relatively few parameters and strong stability. | Local semantic modeling in shallow and deep text feature extraction |
Long Short-Term Memory (LSTM) | Capture long-distance dependencies and global semantic relationships of text | It solves the traditional RNN gradient vanishing and is suitable for long text sequence processing | Deep feature extraction strengthens the semantic coherence between sentence sequences |
Hierarchical neural network | Hierarchical modeling of the local and global structure of text | It preserves text hierarchies and enhances topic-related semantic understanding | In the topic-related feature extraction stage, the relationship between the topic and the topic of the composition is processed. |
Attention mechanism | Dynamically adjust the weights of different features to focus on key semantic information | It enhances the model’s ability to identify important information and improves task adaptability | The theme-related stage supports the multi-task feature fusion of the cross-attention mechanism. |
