一种基于图卷积网络的文本多标签学习方法

多标签学习广泛应用于文本分类、标签推荐、主题标注等.最近,基于深度学习技术的多标签学习受到广泛关注,针对如何在多标签学习中有效挖掘并利用高阶标签关系的问题,提出一种基于图卷积网络探究标签高阶关系的模型TMLLGCN.该模型采用GCN的映射函数从数据驱动的标签表示中生成对象分类器挖掘标签高阶关系.首先,采用深度学习方法提取文本特征,然后以数据驱动方式获得基础标签关联表示矩阵,为更好地建模高阶关系及提高模型效果,在基础标签关联表示矩阵上考虑未标记标签集对已知标签集的影响进行标签补全,并以此相关性矩阵指导GCN中标签节点之间的信息传播,最后将提取的文本特征应用到学习高阶标签关系的图卷积网络分类器进行端到端训练,综合标签关联和特征信息作为最终的预测结果.在实际多标签数据集上的实验结果表明,提出的模型能够有效建模标签高阶关系且提升了多标签学习的效果.

Text Multi-label Learning Method Based on Graph Convolutional Networks

Multi-label learning is widely used in text categorization,label recommendation and topic labeling.Recently,multi-label learning based on deep learning techniques has received widespread attention.In order to explore and exploit high-order label correlations in multi-label learning,a model TMLLGCN which is based on graph convolutional networks(GCN) is proposed.The GCN mapping function generates object classifiers from data-driven label representation to mine label high-order correlations.Firstly,text features are extracted based on deep learning method,the basic label correlation matrix is obtained by data-driven method.Secondly,the unmarked label set is used to perform label completion to improve the effect.After that,the new label correlation matrix is designed to guide information propagation between label nodes in GCN.Finally features extracted from the text are applied to the graph convolutional network classifier for end-to-end training,and the final prediction is made by integrating the label correlation and feature information.The experimental results on the multi-label dataset illustrate the effectiveness of the proposed model.