摘要
针对现有网络安全文本按照主题分类,导致主题之间联系不紧密,文本资源缺乏系统性的问题。论文提出在网络安全知识体系下,按照知识点对文本进行分类。同时,提出Albert-TextCNN模型进行知识点提取,使用Albert模型提取输入文本的词向量,在减小参数量的前提下,输出了包含丰富语义的词向量,并使用TextCNN进一步提取词向量的局部语义特征,最后辅以网络安全知识体系,根据知识点标签之间的包含关系,进一步提升了模型性能。实验结果表明,该模型各项指标均高于实现了同类型单个模型的性能。
The popular classification for cyber security texts is based on their themes,which leads to the lack of connection among the theme labels and the distribution of text unsystematic.This paper presents a method based on Albert-TextCNN to extract the knowledge labels from cyber security texts in the cyber security knowledge framework,which also can make a classification by the knowledge labels.On the premise of reducing the number of parameters,the word embedding of the texts are output,which contains semantic information.Then,TextCNN model can extract the local semantic characteristics.Finally,the cyber security knowledge framework is utilized to obtain the relationship among the knowledge labels as prior knowledge,which can further enhance the model performance.According to various indicators in the experimental results,the model is superior to the same kind of a single model.
作者
刘江舟
段立
LIU Jiangzhou;DUAN Li(Naval University of Engineering,Wuhan 430000)
出处
《舰船电子工程》
2022年第3期114-118,170,共6页
Ship Electronic Engineering
基金
国家自然科学基金项目(编号:41774021)资助。
作者简介
刘江舟,男,硕士研究生,研究方向:自然语言处理;段立,男,博士,教授,研究方向:军事知识工程、信息融合。
