多状态图神经网络文本分类算法

为了提高模型在文本分类任务中的分类性能，针对图神经网络中存在的过度平滑问题，同时增强模型在处理文本特征与文本表示方面的能力，提出了一种基于多状态图神经网络的文本分类算法(multi-state graph neural network, MSGNN)。多状态图神经网络是利用网络层的多个历史状态信息对图神经网络进行强化，构建合理的文本图结构数据作为模型输入。在缓解网络层过度平滑问题的同时，结合2种改进后的不同类型的图神经网络来增强模型的特征提取与特征聚合能力。利用多头自注意力机制对文本关键词的挖掘与利用能力，从多个文本子空间来生成高质量的文本表示，进而完成文本分类。通过在几个公开的文本分类数据集上进行实验分析，相较于其他神经网络的文本分类算法，该方法取得了较好的分类准确率。

Multi-state graph neural network for text classification

To improve the classification performance of the model in the text classification task, aiming at the problem of over-smoothing in the graph neural network and enhancing the ability of the model in processing text features and text representation, we propose a text classification algorithm based on multi-state graph neural network (MSGNN). MSGNN uses multiple historical state information of the network layer to strengthen the graph neural network and constructs reasonable text graph structure data as model input. While the over-smoothing problem of the network layer is being alleviated, two improved graph neural networks of different types are combined to enhance the feature extraction and feature aggregation capabilities of the model. The multi-head self-attention mechanism is used to mine and utilize text keywords. High-quality text representations are generated from multiple text subspaces, and then text classification is completed. Through experimental analysis of several public text classification datasets, the proposed algorithm achieves better classification accuracy than other text classification algorithms based on neural networks.