Dialogue classification method and system based on heterogeneous graph neural network

The invention discloses a dialogue classification method and system based on a heterogeneous graph neural network, and the method comprises the steps: data preparation: collecting a dialogue data set, and dividing a dialogue into sentence nodes and word nodes; constructing a heterogeneous graph according to the dialogue data set; a hidden state is initialized for each node, rich semantic information is captured using BERT-based embedding, and node features are used as initial representation vectors. Constructing a heterogeneous graph neural network model; updating the nodes; and carrying out dialogue classification. The method has the advantages that the method has the capability of capturing numerous-difference language modes, context dependency relationships and user intentions which continuously change across dialogues, and the accuracy and effect of dialogue classification are improved. 本发明公开了一种基于异构图神经网络的对话分类方法和系统，包括：数据准备：收集对话数据集，并将对话划分为句子节点和词节点。根据对话数据集构建异构图；为每个节点初始化隐藏状态，使用基于BERT的嵌入来捕获丰富的语义信息，使用节点特征作为初始表示