摘要
针对图嵌入式文本分类方法在预测性能和归纳能力方面的缺陷,在文本图卷积网络(TextGCN)的基础上,进行适当改进。结合预测文本嵌入(PTE)的高效训练和归纳性,在各个网络层中使用不同的图;通过异质图卷积网络架构来学习特征嵌入,利用习得的特征进行归纳推理。实验结果表明,在大量训练样本标注的情况下,所提方法取得了与其它方法相当或稍优的性能。在少量训练样本标注的情况下,所提方法表现更优,性能增益范围为2%~7%,支持更快的训练和泛化性。
Aiming at the defects of graph embedded text classification method in predictive performance and inductive ability,an appropriate improvement was made on the basis of text graph convolution network(TextGCN).Combined with the efficient training and induction of predictive text embedding(PTE),different graphs were used in each network layer.The feature embedding was learned through the heterogeneous graph convolution network architecture,and the learned feature was used for inductive reasoning.Experimental results show that the proposed method achieves equivalent or slightly better performance than other methods when a large number of training samples are labeled.In the case of a small number of labeled training samples,compared with other excellent methods,the proposed method performs better,and the performance gain range is 2%~7%,supporting faster training and generalization.
作者
赵钦
郑成博
ZHAO Qin;ZHENG Cheng-bo(Department of Computer Science and Technology,Taiyuan University,Taiyuan 030032,China;School of Metallurgy and Energy,North China University of Science and Technology,Tangshan 063000,China)
出处
《计算机工程与设计》
北大核心
2023年第4期1144-1150,共7页
Computer Engineering and Design
基金
计算智能与中文信息处理教育部重点实验室基金项目(2017006)
山西省教育厅山西省高等学校大学生创新创业基金项目(2020703)。
关键词
文本分类
预测性能
文本图卷积网络
异质图卷积网络
预测文本嵌入
归纳推理
特征嵌入
text classification
predictive performance
text graph convolution network
heterogeneous graph convolution network
predictive text embedding
inductive reasoning
feature embedding
