智能与分布计算实验室
  金融交易网络中隐组检测技术研究
姓名 胡广浩
论文答辩日期 2008.06.05
论文提交日期 2008.06.10
论文级别 硕士
中文题名 金融交易网络中隐组检测技术研究
英文题名 Research on Hidden Group Detection technology in Financial Transaction Network
导师1 李玉华
导师2
中文关键词 link mining;hidden group;Hidden Markov Model;Genetic Algorithm
英文关键词 链接挖掘;隐组;隐马尔可夫模型;遗传算法
中文文摘 在复杂金融交易背景下,洗钱是一项重要的金融犯罪活动。随着科技的进步,这种洗钱活动也越来越趋于群组化。这些群组为了隐瞒其犯罪行为,会通过各种手段隐藏其存在,为此将它们称为隐组。 链接挖掘技术不同于以单独的对象作为实例的传统数据挖掘技术,它更加注重对象之间的链接,在金融领域,这种链接即交易也构成了金融数据的主体。基于链接挖掘思想,根据隐组的表现形式,采用隐马尔可夫模型作为金融交易网络的进化模型。金融交易网络中节点的社会结构视为一马尔可夫链,节点之间的交易是互相独立的,且某时刻的交易图仅与该时刻节点的社会结构有关,所以节点产生的交易序列就是符合两个概率的隐马尔可夫过程,这样就建立了金融交易网络的隐马尔可夫模型。由于本文所研究的是隐组问题,因此对不存在隐组的情况不进行深入分析。仅在隐组存在的情况,一方面在确定节点社会结构的情况下,通过模型产生金融交易序列作为算法的输入来验证模型的正确性和有效性。另一方面根据模型产生的金融交易序列作为算法的输入,研究能判定网络存在隐组的概率与所观察交易强度序列周期数之间的关系。 在检测隐组的过程中,遍历解空间的规模随着节点数的增加而成指数增长。为此采用遗传算法作为隐组的求解算法,并结合应用背景,深入讨论了运用遗传算法对其求解的过程,通过实验及结果分析,验证算法的有效性。 设计并开发了金融交易网络中隐组检测的原型系统,给出了模型的整体框图以及主要模块的具体设计和实现流程,并分析了系统的性能。
英文文摘 Under the complex financial background, money-laundering is an important financial criminal activity. Along with the technical progress, there are more and more groups that make money-laundering activities. The group, namely hidden group in this article, tries to hide its existence through kinds of methods to conceal its criminality. Link mining technology is different from traditional data mining technology that treats all the objects as independent ones and it even more pays great attention to the links which constitute the main data in the financial domain. This article considers the hidden group’s behavior, and takes the Hidden Markov Model as the financial transaction network’s evolutional model. In the financial transaction network the nodes’ social structure is a Markov chain. The transactions between two nodes are mutually independent, and the transaction graph is only determined by the social structure at some time, therefore the transaction sequence conforms to the Hidden Markov Processes. This article do not pay attention to the situation that exists no groups because we are only interested in hidden group. On the one hand, we present a social structure and input the financial transaction sequence from the Hidden Markov Model. We observe the model’s accuracy and validity. On the other hand, this research focuses on the relation between the probability that we think a hidden group exists and the transaction intensity sequence periodicity. When we detect the hidden group, the space’s scale we must visit increases exponentially along with the nodes’ growth. Therefore this article uses the Genetic Algorithm as the hidden group detection’s algorithm, and discusses the solution process under the application background, and analyzes the results of the experiment to confirmate the algorithm’s validity. An archetypal system for group detection in financial transaction network is designed and developed. Introduces components of the model, main designing and work flow. Moreover, performance analysis of the system is presented.