【作者】 董旭；

【导师】 张宏莉；

【作者基本信息】 哈尔滨工业大学 ， 网络空间安全， 2019， 硕士

【摘要】 在互联网飞速发展的今天,网络用户规模不断攀升,网络信息数量日益增长,尤其以社交网络为代表的互动性平台,更是紧密连接了每一个现实个体与虚拟的网络空间。由于互联网的各个虚拟主体背后的操控者是现实中的真实个体,通过对互联网中的基本特征的深入挖掘,可以分析网络主体在现实中特定行为的倾向性。社会的不断发展,催生了对公共安全的更高需求,尤其是犯罪预测等方面。在此背景下,本文以犯罪社会学相关学科为依据,分析了大量犯罪个体,研究了一种分析网络主体在现实空间中犯罪倾向的分析方法,为公共安全、刑侦、警务等提供技术支持。本文获取到大量网络主体的相关信息,结合相关学科的研究,归纳出多个主体基本特征,并进行初步挖掘。这些特征,可以唯一标定网络与现实的虚实映射关系,也在一定程度上沟通主体的行为表现。这种同时具有“推断表征”和“标定个体”两大功能的数据集合,与生物学上的“基因”概念异曲同工。因此本文借助网络主体基因理论,将网络基因的研究层次划分为属性、基因单元、基因片段三个层次。基因单元对网络主体行为性状表现起决定作用,本文结合犯罪学、社会学等相关学科的研究,定义了暴力单元、焦虑单元、社会联系单元等多种与犯罪行为相关的基因单元。同时,针对每一种基因单元,本文给出了对应的抽取计算的方式。本文借鉴生物学在基因研究中的研究思路,使用非等位基因的类别命名,按照基因表达效果对基因单元进行分类,对于每一种类别,本文给出了定义、入选规则和计算方法。在行为倾向计算时使用这种分类过的基因算子,降低了特征维度,减小了机器学习的误差;同时也使计算过程拥有了一定的健壮性、可扩展性和通配性,为未来本研究的深入探索和社交网络其他相关研究提供了一种新的研究方法。本文以网络主体基因作为研究基础,以犯罪为代表的特定行为作为研究目的,借助网络主体基因,构建了一种基于网络主体的特定行为倾向分析算法CSCTA。算法主要分为三个步骤,第一步是进行基因算子的内部数据计算,第二步是使用XGBoost等方法计算基因表达强度,第三步是犯罪倾向性判别算法。使用本文的方法,针对犯罪这一特定行为分析,准确度达到了79%,精准度达到了75%。与相关研究常用的方法进行比较,本文给出的算法精准度更高,特征挖掘也取得了成效,证明了本文提出的CSCTA算具有良好的效果,整体达到了可用的标准。更多还原

【Abstract】 Nowadays,along with the rapid development of the Internet,the size of Internet users and the amount of online information have also been rising with each passing day.In particular,the social network – represented interactive platforms have closely connected the virtual cyber space with each individual in the reality.Since the controllers behind each virtual subject on the Internet were individuals in reality,an in-depth exploration of the basic characteristics of the Internet could serve as a reference to the tendency of specific acts of cyber subjects in the real world.The continuous development of the society has given birth to higher demands for public security,especially in crime prediction and other aspects.In such a context,with Criminal Sociology – related disciplines as the basis,the author analyzed a large number of individuals who have committed crimes and developed an analytical method to evaluate cyber subjects’ criminal intent in the real world,thus providing technical support to public security,criminal investigation,policing,etc.This paper collected related information of a large number of cyber subjects.On such a basis,combined with studies of relevant disciplines,the author concluded basic characteristics of multiple subjects and completed preliminary exploration.These characteristics could uniquely calibrate the virtual – real mapping between cyber and the reality,and could communicate behavioral performance of the subjects to a certain extent.Data sets with both “characterization inferring” and “individual calibrating” functions achieved the same goal with the biological concept “gene”,only by different means.Therefore,in this paper,referring to the theory of cyber subject gene,the author divided the research of cyber gen into three levels,namely property,genetic element and gene fragment.Genetic elements were of a decisive role in the expression of cyber subjects’ behavioral characters.In this paper,based on studies in criminology,sociology and other related disciplines,the author defined the violent element,the anxious element,the social bond element and other gene elements of various kinds related to criminal behavior.Meanwhile,targeting each gene element,the author put forward the corresponding extraction and calculation methods.Drawing on the biological research ideas with respect to genes,in this paper,the author introduced the names of categories of non-allelic genes and categorized gene elements by the effect of genetic expression.Each category was defined and provided with rules of inclusion and computational methods.Upon calculation of behavioral tendencies,the use of such categorized gene elements could reduce characteristic dimensions and the error in machine learning;in the meantime,it equipped the computational procedure with certain robustness,scalability and compatibility,thus bringing forward a new research method for future exploration in this regard and other related research in the field of social network.With cyber subject genes as the base of research and specific acts represented by crime committing as the objective,in this paper,the author established a cyber subject – based analyzing algorithm of their specific action tendencies,CSCTA,with the help of the theory of cyber subject genes.The algorithm was mainly divided into three steps: first,calculation of the internal data of genetic operators;second,calculation of the strength of genetic expression with XGBoost and other methods;third,the algorithm to distinguish criminal tendencies.The method herein achieved the accuracy of 79% and the precision of 75% in the analysis of the specific behavior of crime committing.Comparison with commonly used methods in related studies indicated that,the algorithm herein was of higher accuracy.Characteristic exploration was also proved effective,verifying the satisfactory performance and overall usability of CSCTA proposed in this paper.更多还原