| 陈浩,王兴芬.基于神经网络的嵌套命名实体关系抽取模型[J].情报工程,2021,7(6):067-0755 |
| 基于神经网络的嵌套命名实体关系抽取模型 |
| Nested Named Entity Recognition Based on Neural Network |
| |
| DOI：10.3772/j.issn.2095-915X.2021.06.006 |
| 中文关键词: 深度学习；嵌套实体识别；实体关系抽取；威胁情报； |
| 英文关键词: Deep learning; nested entity recognition; entity relationship extraction; threat intelligence |
| 基金项目:国家重点研发计划课题“大宗商品电子商务市场的交易风险智能分析与预警技术”（2019YFB1405003）；北京市教委科技一般项目“数字化转型背景下面向企业协同治理的群体智能决策机制与方法研究”（KM202111232017）。 |
| 作者 | 单位 | | 陈浩 | 北京信息科技大学 北京 100192 | | 王兴芬 | 北京信息科技大学 北京 100192 |
|
| 摘要点击次数: 2000 |
| 全文下载次数: 1692 |
| 中文摘要: |
| [ 目的/ 意义] 网络安全形势日益严峻，从威胁情报中抽取网络安全实体及其关系，构建结构化威胁情报信息，对于网络安全防护来说尤为重要。在过去的工作中嵌套实体的关系抽取一直是难点，嵌套实体和关系重叠不能被有效识别导致准确率低，为威胁情报信息抽取研究带来巨大的挑战。[ 方法/ 过程] 基于上述情况，针对嵌套实体关系抽取过程中存在的问题，本文基于图注意力网络提出了一种新型知识抽取模型SRG。采用Bert-Bi-LSTM 作为共享编码层，与边界检测模块所得的多跨度实体共同进行跨度表示后，利用图注意力网络提取特征进行关系抽取，可有效改善实体边界信息与类别信息的检测效果。[ 结果/ 结论] 在公共数据集上进行了实验验证，验证结果表明，其在解决实体嵌套与关系重叠的问题上有显著的效果。 |
| 英文摘要: |
| [Objective/Significance]The network security situation is becoming much more serious. It is particularly important for network security protection to extract network security entities and their relationships from threat intelligence and build structured threat intelligence information. In the past work, the relationship extraction of nested entities are always difficult. The overlapping of nested entities and relationships cannot be effectively identified, resulting in low accuracy, which brings great challenges to the research of threat intelligence information extraction. [Methods/Process]According to such situation and aiming at the problems existing in the process of nested entity relationship extraction, this paper proposes a new knowledge extraction model SRG based on graph attention network. Bert-Bi-LSTM is used as the shared coding layer to represent the span together with the multi span entities obtained by the boundary detection module, and the graph attention network is used to extract features for relationship extraction, which can effectively improve the detection effect of entity boundary information and category information. The experimental verification is carried out on the public data set. [Results/Conclusions]The verification results show that, it has a significant effect on solving the problem of entity nesting and relationship overlap. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
