面向小样本的威胁情报命名实体识别方法  

Named entity recognition method for thread intelligence for small samples

作　　者：萨仁高娃 邬超慧 张振 张悦 Sa Ren Gaowa;WU Chao-hui;ZHANG Zhen;ZHANG Yue(Inner Mongolia Economic and Technological Research Institute,Inner Mongolia Electric Power Group Co.,Ltd,Hohhot 010020,China)

机构地区：[1]内蒙古电力集团有限责任公司内蒙古电力经济技术研究院,内蒙古呼和浩特010020

出　　处：《计算机工程与设计》2024年第9期2599-2605,共7页Computer Engineering and Design

基　　金：国家自然科学基金项目(59637050)。

摘　　要：为解决威胁情报领域的数据源不足、实体专业性强等问题,提出一种面向小样本的威胁情报命名实体识别模型AbNER。设计基于自注意力机制的隐式特征引导模块,引入prompt模板,融合专业领域的先验知识,结合两者共同完成识别实体。对模型输入层结构进行优化,有效提升编码性能。分析AbNER在通用和电网等两类威胁情报数据上的测试结果,模型在5个全量数据集和3个小样本数据集上均达到最优表现,验证了AbNER的实体识别优势和小样本能力。To solve the problems of insufficient data sources and strong entity professionalism in the field of threat intelligence,AbNER,a named entity recognition model for threat intelligence for small samples was proposed.An implicit feature guidance module based on the self-attention mechanism was designed,a prompt template was introduced to integrate the prior knowledge of the professional field,and the two were combined to jointly complete the identification of entities.The model input layer structure was optimized to effectively improve the coding performance.The test results of AbNER on two types of threat intelligence data,such as general and power grid,were analyzed.The model achieves the best performance on 5 full data sets and 3 small sample data sets,which verifies the entity recognition advantages and small sample capabilities of AbNER.

关 键 词：命名实体识别 威胁情报 小样本 自注意力机制 大规模语言模型 提示学习 网络安全 

分 类 号：TP389.1[自动化与计算机技术—计算机系统结构]