基于标签构建与特征融合的多标签文本分类研究方法  

Research methodology of multi-label text classification based onlabel construction and feature fusion

作　　者：王旭阳[1] 卢世红 WANG Xuyang;LU Shihong(School of Computer and Communication,Lanzhou University of Technology,Lanzhou,Gansu 730050,China)

机构地区：[1]兰州理工大学计算机与通信学院,甘肃兰州730050

出　　处：《贵州师范大学学报(自然科学版)》2025年第1期105-114,共10页Journal of Guizhou Normal University:Natural Sciences

基　　金：国家自然科学基金项目(62161019)。

摘　　要：目前存在的多标签文本分类任务算法,对于标签的建模不是很成熟,其中对于标签的依赖性问题,以及标签特征和文本特征的融合程度问题,均缺乏有效的处理方法。为了更有效地利用标签间的依赖关系,以及整合标签特征与文本特征的融合,提出了一种名为CGTCN的多标签文本分类模型。该模型从标签构建和特征融合的角度出发,通过CompGCN建模标签依赖关系,先利用Transformer中的多头交叉注意力机制初步融合标签特征和文本特征,然后再通过CorNet网络进一步捕获标签特征与文本特征之间的相关性,从而得到最终的标签预测。实验结果显示,与基准模型相比,该方法能够有效的提升模型性能,在多标签文本分类任务中取得更好的分类效果。Currently,the existing algorithms for multi-label text classification tasks are not very mature in modeling labels.They lack effective methods to address issues such as label dependencies and the degree of integration between label features and text features.In order to more effectively utilize dependencies among labels and integrate label features with text features,a multi-label text classification model named CGTCN is proposed.This model approaches label construction and feature integration by first modeling label dependencies by using CompGCN.It initially integrates label features and text features through the multi-head cross-attention mechanism of Transformer,and then further captures the correlation between label features and text features by using the CorNet network to obtain the final label predictions.Experimental results demonstrate that compared to baseline models,this method effectively improves model performance and achieves better classification results in multi-label text classification tasks.

关 键 词：多标签文本分类 CompGCN TRANSFORMER CorNet 标签相关性 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]