面向用户偏好发现的隐变量模型构建与推理

发布时间：2018-02-16 09:12

本文关键词： 用户偏好评分数据贝叶斯网隐变量模型概率推理带偏置的矩阵分解　出处：《计算机应用》2017年02期 　论文类型：期刊论文

【摘要】：电子商务应用中产生了大量用户评分数据,而这些数据中富含了用户观点和偏好信息,为了能够从这些数据中准确地推断出用户偏好,提出一种面向评分数据中用户偏好发现的隐变量模型(即含隐变量的贝叶斯网)构建和推理的方法。首先,针对评分数据的稀疏性,使用带偏置的矩阵分解(BMF)模型对其进行填补;其次,用隐变量表示用户偏好,给出了基于互信息(MI)、最大半团和期望最大化(EM)算法的隐变量模型构建方法;最后,给出了基于Gibbs采样的隐变量模型概率推理和用户偏好发现方法。实验结果表明,与协同过滤的方法相比,该方法能有效地描述评分数据中相关属性之间的依赖关系及其不确定性,从而能够更准确地推断出用户偏好。
[Abstract]:A large number of user rating data are generated in e-commerce applications, which are rich in user opinion and preference information, in order to accurately infer user preferences from these data. In this paper, a method of constructing and reasoning hidden variable model (Bayesian network with hidden variables) for user preference discovery in scoring data is proposed. Firstly, aiming at the sparsity of score data, The matrix factorization (BMF) model with bias is used to fill it. Secondly, implicit variables are used to represent user preference, and a method of constructing hidden variable model based on mutual information MIA algorithm, maximum half-group algorithm and expectation maximization algorithm is presented. The probabilistic reasoning and user preference discovery method of hidden variable model based on Gibbs sampling are presented. The experimental results show that, compared with the cooperative filtering method, This method can effectively describe the dependency and uncertainty among the related attributes in the scoring data, and thus more accurately infer user preferences.
【作者单位】：云南大学信息学院;昆明理工大学信息工程与自动化学院;
【基金】：国家自然科学基金资助项目(61472345,61562090,61462056) 云南省应用基础研究计划项目(2014FA023,2014FA028) 云南省中青年学术和技术带头人才后备人才培育计划项目(2012HB004) 云南大学青年英才培育计划项目(XT412003);云南大学创新团队培育计划项目(XT412011)~~
【分类号】：F713.36;TP18

【相似文献】