基于贝叶斯网的评价数据分析和动态行为建模

发布时间：2018-03-04 02:04

本文选题：用户评价数据　切入点：时序性　出处：《计算机研究与发展》2017年07期 　论文类型：期刊论文

【摘要】：随着Web2.0的不断普及和电子商务应用的迅速发展,大规模的在线评价数据不断产生,使用户行为数据分析和用户行为建模成为可能,具有重要意义.考虑到用户评价数据和评价行为的动态性,提出以带有隐变量的贝叶斯网作为各属性间依赖关系及其不确定性表示的基本框架,构建既能刻画用户评价数据中各属性间相互依赖的不确定性、也能描述用户行为动态性的评价行为模型.首先,以贝叶斯信息标准(BIC)分值作为模型与数据拟合度的度量标准,提出基于打分搜索方法来构建各时间片的隐变量模型,并给出基于期望最大(EM)算法的隐变量取值填充方法;其次,基于条件互信息和时序的不可逆性,提出了相邻时间片间隐变量模型的构建方法.建立在MovieLens数据集上的实验结果验证了提出的动态用户行为建模方法的高效性及有效性.
[Abstract]:With the popularity of Web2.0 and the rapid development of e-commerce applications, large-scale online evaluation data are constantly produced, which makes user behavior data analysis and user behavior modeling possible. Considering the dynamic nature of user evaluation data and evaluation behavior, a Bayesian network with hidden variables is proposed as the basic framework for the representation of dependencies and uncertainties among attributes. This paper constructs an evaluation behavior model which not only depicts the uncertainty of interdependence among attributes in user evaluation data, but also describes the dynamics of user behavior. Firstly, the Bayesian information standard (BIC-score) is used as the metric of the fitting between the model and the data. The hidden variable model of each time slice is constructed based on scoring search method, and the filling method of hidden variable value based on the expected maximum EM-based algorithm is given. Secondly, based on conditional mutual information and irreversibility of time series, the implicit variable filling method is proposed. A method of constructing the hidden variable model between adjacent time slices is proposed, and the experimental results based on the MovieLens data set verify the efficiency and effectiveness of the proposed dynamic user behavior modeling method.
【作者单位】：云南大学信息学院;云南大学科技处;
【基金】：国家自然科学基金项目(61472345,61562090) 云南省应用基础研究计划重点项目(2014FA023) 云南大学青年英才培育计划项目(WX173602);云南大学创新团队培育计划项目(XT412011) 云南省教育厅科研基金项目(2016ZZX006)~~
【分类号】：F713.36;TP18

【参考文献】