社会网络中用户行为分析及预测研究

发布时间：2018-03-17 03:29

  本文选题：转发行为预测　切入点：重叠社交圈发现　出处：《吉林大学》2016年博士论文　论文类型：学位论文

【摘要】：影响信息传播过程的关键因素之一是网络结构,由于用户是社会网络中的行动主体,因此,用户行为亦影响着信息的传播过程,有效地分析并挖掘社会网络中的用户行为背后蕴藏的深层次规律有助于深入理解社会网络的形成及演化机制。本文针对社会网络中用户行为分析及预测的相关问题,展开了深入研究,主要研究内容包括:(1)针对用户转发信息的行为,通过量化用户社会关系因子约束目标函数,将预测问题转化为求解用户概要和用户发布内容两个维度的最优解问题,提出一种基于加权非负矩阵分解的用户转发行为预测算法;(2)针对用户自定义社交圈的行为,通过重新定义密度估计函数和增加社交圈整合步骤,提出一种改进的基于密度点聚类的重叠社交圈发现方法;(3)基于垃圾用户和正常用户不同的行为,通过重新定义亲和力度量标准、多样化的亲和力阈值以及标准正态分布变异因子,提出一种基于改进人工免疫算法的垃圾用户识别模型;(4)基于五大人格特质所表现出的用户行为,通过将动态的用户人格特质阈值以及关于用户人格特质间相关性的先验知识引入多标记分类算法中,提出一种基于改进ML-KNN算法的多维用户人格特质识别模型。在多个公用数据集上的实验表明,本文提出的方法能获得较好的效果。
[Abstract]:One of the key factors affecting the process of information dissemination is the network structure. As users are the main actors in social networks, user behavior also affects the process of information dissemination. It is helpful to understand the formation and evolution mechanism of social network by analyzing and mining the deep-seated law of user behavior in social network. This paper aims at the related problems of user behavior analysis and prediction in social network. The main contents of this study include: 1) quantifying the objective function of user social relationship factor, according to the behavior of transmitting information. In this paper, the prediction problem is transformed into the optimal solution problem of two dimensions of user summary and user content, and a user forwarding behavior prediction algorithm based on weighted non-negative matrix decomposition is proposed, which is aimed at the behavior of user-defined social circle. By redefining the density estimation function and increasing the social circle integration step, an improved social circle discovery method based on density point clustering is proposed, which is based on different behaviors of garbage users and normal users. By redefining affinity metrics, diverse affinity thresholds, and standard normal distribution variation factors, A garbage user identification model based on improved artificial immune algorithm (AIA) is proposed. By introducing dynamic user personality trait threshold and prior knowledge about the correlation between user personality traits into multi-marker classification algorithm, A multi-dimensional user personality recognition model based on improved ML-KNN algorithm is proposed. Experiments on several common data sets show that the proposed method can achieve good results.
【学位授予单位】：吉林大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TP393.09;TP18
，

本文编号：1622985