论文部分内容阅读
用户偏好是决定用户对商品评分的隐含变量,以构建包含用户偏好的隐变量模型、描述评分数据中相关属性间任意形式依赖关系及其不确定性为主要目标,以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,由商品评分数据构建不含隐变量的商品评分模型,提出基于半团结构向其中插入描述用户偏好的隐变量的方法,从而构建包含用户偏好的隐变量模型,并给出基于EM算法的隐变量模型参数估计方法,进而提出隐变量模型的概率推理算法和相应的商品评分预测方法.建立在MovieLens和Book-Crossing数据上的实验结果表明,本文提出的隐变量模型构建和相应的评分预测方法是有效的.
The user preference is the implicit variable that determines the user’s rating of the product, so as to construct a latent variable model containing the user’s preference, describe the arbitrary dependence between the related attributes in the scoring data and its uncertainty as the main objective, and use the Bayesian network as Attribute dependency relationship and its uncertainty expression, a commodity scoring model without implicit variables is built based on the product scoring data, and a method of inserting implicit variables describing user preferences based on the semigroup structure is proposed to construct a scoring model that contains user preferences , And gives the method of parameter estimation of hidden variable model based on EM algorithm, and then puts forward the probabilistic reasoning algorithm of hidden variable model and the corresponding commodity score forecasting method.Experimental results based on MovieLens and Book-Crossing data show that, The proposed latent variable model construction and the corresponding score prediction method is effective.