This paper proposes a novel Graph Convolutional Network (GCN) model with dense connections and an attention mechanism for text classification. The model, called DC-GCN, addresses the limitations of existing GCN models by increasing the depth of the network and using dense connections to capture long-distance node information and various scale features of text. An attention mechanism is also introduced to combine features and determine their relative importance. The model is evaluated on four benchmark datasets and shows significantly higher classification accuracy compared to conventional deep learning text classification models and other GCN-based text classification methods. The model's key contributions include a new way to build a separate graph for each sentence, extending dense connections to GCN networks, and using an attention mechanism to automatically determine the relative relevance of various word nodes. The model outperforms existing methods on short and medium text classification tasks but shows weaker performance on long text classification tasks. The results demonstrate the effectiveness of graph convolution in text feature extraction and the model's ability to adaptively extract text features at various scales. The paper also discusses the influence of network depth and window size on classification accuracy and explores the effectiveness of the attention mechanism. Overall, the proposed model shows significant improvements in text classification performance, particularly for short and medium texts.This paper proposes a novel Graph Convolutional Network (GCN) model with dense connections and an attention mechanism for text classification. The model, called DC-GCN, addresses the limitations of existing GCN models by increasing the depth of the network and using dense connections to capture long-distance node information and various scale features of text. An attention mechanism is also introduced to combine features and determine their relative importance. The model is evaluated on four benchmark datasets and shows significantly higher classification accuracy compared to conventional deep learning text classification models and other GCN-based text classification methods. The model's key contributions include a new way to build a separate graph for each sentence, extending dense connections to GCN networks, and using an attention mechanism to automatically determine the relative relevance of various word nodes. The model outperforms existing methods on short and medium text classification tasks but shows weaker performance on long text classification tasks. The results demonstrate the effectiveness of graph convolution in text feature extraction and the model's ability to adaptively extract text features at various scales. The paper also discusses the influence of network depth and window size on classification accuracy and explores the effectiveness of the attention mechanism. Overall, the proposed model shows significant improvements in text classification performance, particularly for short and medium texts.