摘要
为预测恐怖袭击事件的嫌疑组织,选取全球恐怖主义数据库;基于恐怖袭击事件发生的时间特性,采用2014—2016年发生的恐怖袭击数据作为训练集,对2017年发生的恐怖袭击事件的发动组织进行分类预测。采用综合采样技术平衡训练集数据,运用双向循环神经网络学习数据集的时间特性,结合自注意力机制,构建基于自注意力机制的双向门控循环神经网络组合模型,对恐怖袭击事件的犯罪嫌疑组织进行分类预测,并将该模型与引入注意力机制的神经网络模型进行对比。研究表明,该模型在预测恐怖袭击事件的犯罪嫌疑组织上具有更高的分类精度,能够为警方快速侦破恐怖袭击案件提供有价值的信息。
Terrorist attacks occur frequently all over the world,which give rise to great harm to the public,and the anti-terrorism situation is extremely urgent.Against this backdrop,it is increasingly vital to track down a criminal quickly and carry out immediate action.With the method of data mining,this paper focuses on exploring the characteristics of terrorist activities and predicting suspected criminal organizations,which can assist the police in tracking down the criminal much quicker.The data used in this paper come from the GTD database of global terrorist attacks,ranging from 1970 to 2017.Additionally,based on the time characteristics of terrorist attacks,the terrorist attack data from 2014 to 2016 are used as the training set to classify and predict the launching organizations of terrorist attacks in 2017.First,this paper constructs a terrorist attack index system by selecting variables and processing the data set,including removing missing values and feature coding.Then,due to the large difference in the number of samples of different terrorist organizations in the data set,to address this problem,a comprehensive sampling method(Synthetic Minority Oversampling Technique and Tomek Link,SMOTE T)is applied to balance the training set.After that,combined with the self-attention mechanism,the Bidirectional Gated Recurrent Unit(BiGRU)is used to learn the time characteristics of the data set,and the three-fold cross-validation random search method is used to determine the parameters of the network.The BiGRU base on Self-Attention(BiGRU SA)neural network combination model is constructed to classify and predict the suspected criminal organizations of terrorist attacks.Finally,the model is compared with other models introducing an attention mechanism(BP,RNN,LSTM,GRU).The results show that the prediction accuracy and macro F1 on the testing set of the BiGRU SA model proposed in this paper are 60.60%and 49.53%,respectively,which is better in predicting the suspected criminal organizations of terrorist attacks than other m
作者
姜旭初
吴沁珏
JIANG Xuchu;WU Qinjue(School of Statistics and Mathematics,Zhongnan University of Economics and Law,Wuhan 430073,China)
出处
《安全与环境学报》
CAS
CSCD
北大核心
2023年第6期2017-2023,共7页
Journal of Safety and Environment
关键词
公共安全
嫌疑组织预测
双向循环神经网络
自注意力机制
综合采样技术
public safety
suspected organization prediction
bidirectional recurrent neural network
self-attention mechanism
integrated sampling technology
