Application Of Hash Feature Extraction And Depth Self-encoding Representation In Counter-terrorism

The development of economic globalization has also made the terrorist attacks show the spread of globalization.It not only undermines the good order of peace and stability of the international community,but also seriously affects the safety of human life and property.Fighting terrorist attacks has become an urgent task.How to study the essential feature and evolution of terrorist attacks and the terrorist attacks that detect a series of unknown murderers have become the focus of more and more countries.In order to study the underlying laws behind the massive information on terrorist attacks,more and more intelligence analysis departments and anti-terrorist law enforcement personnel have studied the transformation of unstructured data related to terrorist attacks into structured data,and then use data mining techniques to extract valuable information from massive terrorist attacks.It will provide intelligence support for decision-making in the counter-terrorism department.In order to find out the potential gangs of terrorist attacks in unknown murderers and assist counter-terrorism intelligence analysts to detect cases,this article uses the Global Terrorism Database(GTD).According to the original data set of terrorist attacks,after data pre-processing,the “Hash Key Information Extraction Method” is used to analyze the terrorist attack data,and select a subset of features that better reflect the true distribution and essential characteristics of terrorist attacks,greatly reducing the dimension of the data,reducing the 133-dimensional from the original data set to 32-dimensional,and using "0" or "1" to indicate the value of each attribute,"0" means that the attribute has little effect on the terrorist attack," 1" represents the influence of this attribute on terrorist attacks.In this paper,the “Hash Key Information Extraction Method” not only greatly reduces the data to be analyzed,but also improves the training and detection speed,and also effectively reduces the noise data,so that it can more effectively reflect the characteristic properties of the relevant data.The experimental results of three clustering algorithms,K-Means,BIRCH and GMM,prove the validity of the “Hash Key Information Extraction Method”.Furthermore,based on the selected feature subsets,this paper proposes an improved clustering algorithm based on deep self-coding representation.Firstly,it maps the sparse and under-standard raw data into compact and smooth data within the class,and then performs cluster analysis.According to the experimental verification of three classical clustering algorithms,the improved clustering algorithm based on deep self-coding representation has improved the clustering accuracy,silhouette coefficient and Calinski-Harabasz compared with the traditional clustering algorithm.Finally,based on the clustering of data,this paper uses a semi-supervised learning method to label unlabeled terrorist attacks in a probabilistic manner to explore the unlabeled terrorist attacks and the labeled terrorist attacks merge.In turn,it provides decision support for the deployment of actions related to the counter-terrorism department,and provides data support for the detection of similar cases and the improvement of the efficiency of the case.The hash feature extraction and depth self-coding representation methods proposed in this paper have certain versatility and expansibility,which can provide reference for data feature extraction in other research fields.The proposed clustering framework can also be extended to other research fields.