摘要
传统的多标签文本分类算法往往只关注文本本身的信息,在挖掘标签间的关联信息过程中存在不足,很大程度上影响了模型的分类性能。为了充分利用标签信息捕获标签之间的依赖关系,提出一种融合注意力与CorNet的多标签文本分类算法。该算法针对文本内容编码部分,使用双向长短时记忆网络获得文本语义表示,通过多标签注意力机制捕获文本中的重要单词;同时,针对标签编码部分,利用图注意力网络捕获标签间的依赖关系,设计“文本-标签”注意力机制将其与文本上下文语义信息进行交互,获得基于标签语义信息的文本特征表示;最后,使用自适应融合机制将上述两部分融合,并通过CorNet模块学习标签相关性以增强标签预测。在AAPD、RCV1-V2和Reuters-21578数据集上的实验结果表明,所提出的算法能够有效捕获标签之间的依赖关系,其性能优于当前主流的多标签文本分类算法。
Traditional multi-label text classification algorithms usually only focus on text information,and there are shortcomings in the process of mining tag-related information,which largely reduces the classification performance of the model.In order to make full use of the dependency between labels captured by tag information,a multi-label text classification algorithm combining attention with Correlation Networks(CorNet)is presented.First,in the text content encoding section,the algorithm uses a two-way long-term and short-term memory network to get the semantic representation of the text,and to capture important words in the text through a multi-label attention mechanism.At the same time,for the label encoding part,the attention network is used to capture the dependencies between labels,the text-label attention mechanism is designed to interact with the text context semantic information to get the text feature representation based on the tag semantic information.Finally,an adaptive fusion mechanism is used to fuse the above two parts,and CorNet module is used to learn label correlation to enhance label prediction.Experiments on AAPD,RCV1-V2,and Reuters-21578 datasets show that the proposed algorithm can capture label dependencies effectively and outperforms the current main multi-label text classification algorithms.
作者
邓维斌
王智莹
高荣壕
王国胤
胡峰
DENG Weibin;WANG Zhiying;GAO Ronghao;WANG Guoyin;HU Feng(Chongqing Key Laboratory of Computational Intelligence,Chongqing University of Posts and Telecommunications,Chongqing 400065,China)
出处
《西北大学学报（自然科学版）》
CAS
CSCD
北大核心
2022年第5期824-833,共10页
Journal of Northwest University（Natural Science Edition）
基金
国家重点研发计划项目(2018YFC0832100)
国家自然科学基金重点项目(61936001)
重庆市自然科学基金创新群体项目(cstc2019jcyj-cxttX0002)
重庆市自然科学基金(cstc2021ycjh-bgzxm0013)
重庆市教委重点合作项目(HZ2021008)。
