面向互联网用户行为分析的加权概率融合贝叶斯网络研究  

Weighted Probabilistic Fusion Bayesian Network for Internet User Behavior Analysis

作　　者：王佳 张文波[1] 朱宏博 WANG Jia;ZHANG Wenbo;ZHU Hongbo(Shenyang Ligong University,Shenyang 110159,China)

机构地区：[1]沈阳理工大学信息科学与工程学院,沈阳110159

出　　处：《沈阳理工大学学报》2023年第4期40-47,共8页Journal of Shenyang Ligong University

基　　金：国家自然科学基金项目(62102272)。

摘　　要：针对依靠单一算法训练互联网用户行为数据构建的贝叶斯网络(Bayesian Network,BN)计算耗时长、结构不稳定等问题,提出加权概率融合并行贝叶斯网络增量学习(WPFPBayes)算法。该算法根据自适应数据切片算法找出最优数据片尺寸,快速进行并行BN模型训练;将数据切片上学习得到的若干子BN结构通过融合加权概率方法合并成一个全局BN模型;通过一种增量评分函数定量表示单位时间内网络模型与数据之间适应程度的变化情况;采用依据特定结点进行BN更新的措施达到新旧数据在网络中的平衡。仿真实验结果表明:WPFPBayes算法下得出的BN模型的效率及其准确率均高于其他常见算法;随着数据量的增加,BN模型数据表达的准确率和稳定性均得到提高,可以更有效检测网络用户的异常行为。To address the problems of computationally time-consuming and unstable structures of Bayesian networks constructed by relying on a single algorithm for training Internet user behaviour data,the Weighted Probabilistic Fusion Parallel Bayesian(WPFPBayes)network incremental learning algorithm is proposed.The algorithm is based on an adaptive data slicing algorithm to find the optimal data slice size for fast training of the parallel Bayesian network model;several sub-Bayesian network structures learned on the data slices are combined into a global Bayesian network model by a fusion weighted probability method;an incremental scoring function is used to quantify the change in the degree of adaptation between the network model and the data per unit time;Bayesian network updates are based on specific nodes to achieve a balance between old and new data in the network.The simulation experiments show that the efficiency and accuracy of the Bayesian network model derived from the WPFPBayes algorithm are higher than those of other common algorithms,and the accuracy and stability of the data representation of the Bayesian network model are improved as the amount of data increases.

关 键 词：加权概率融合贝叶斯网络 贝叶斯网络更新 自适应数据切片 网络用户行为 

分 类 号：TP274[自动化与计算机技术—检测技术与自动化装置]