特征融合的多标签文本分类研究
Research on Multi-label Text Classification Based on Feature Fusion
李楚贞 1江涛1
作者信息
- 1. 广东理工学院信息技术学院,广东肇庆 526100
- 折叠
摘要
作为多标签文本分类的一个重要步骤,目前特征提取方法已取得重大进展,但基于深度学习的特征提取方法存在获取特征单一、不全面等问题,因此,本文提出新的特征融合提取模型,即使用BiGRU提取文本的全局特征,Capsule network提取文本的局部特征和位置信息,同时使用TF-IDF提取文本的统计特征.实验证明该模型在公共数据集RCV1-V2和AAPD上的性能都得到改进.
Abstract
As an important step in multi-label text classification,currently,significant progress has been made in feature extraction methods.Nevertheless,the obstacles of single and incomplete feature acquisition based on deep learning remain.Therefore,this paper proposes a new feature fusion extraction model,which employs BiGRU to extract global features of text,Capsule network to extract local features and position information of text,and TF-IDF to extract statistical features of text.Experiments have shown that the performance of this model has been improved on both the public dataset RCV1-V2 and AAPD.
关键词
多标签/分类/特征融合Key words
multiple labels/classification/feature fusion引用本文复制引用
基金项目
广东理工学院科技项目(2022GKJZK005)
出版年
2024
NETL
NSTL
万方数据