基于弹性网络正则化的隐因子预测模型  

LATENT FACTOR PREDICTION MODEL BASED ON ELASTIC NETWORK REGULARIZATION

作　　者：贺文灏 王德贤[1] 邓萍 刘锐 He Wenhao;Wang Dexian;Deng Ping;Liu Rui(School of Computer Science,China West Normal University,Nanchong 637000,Sichuan,China;College of Information Science and Technology,Southwest Jiaotong University,Chengdu 610000,Sichuan,China)

机构地区：[1]西华师范大学计算机学院,四川南充637000 [2]西南交通大学信息科学与技术学院,四川成都610000

出　　处：《计算机应用与软件》2019年第7期287-293,共7页Computer Applications and Software

基　　金：国家自然科学基金面上项目(61871330);国家大学生创新创业专项基金项目(201510638047,201810638020x);四川省教育厅重点项目(15ZA048,13ZA0015);四川省科技厅支撑项目(2018GFW0151);西华师范大学英才基金资助课题(17YC150,17YC149)

摘　　要：在大数据预测中,通常采用SGD_LF模型对高维稀疏矩阵中的缺失数据进行预测。由于SGD_LF模型仅有L 2正则化项对目标函数进行约束,不能调整隐因子的分布,这样导致模型不能很好地描述目标矩阵中实体特性,损失了模型精度。用FOBOS算法构造一个同时用L 1和L 2限制目标函数的弹性网络ENLF模型,ENLF很好地调整隐因子的分布并提高了模型性能。为了进一步提高ENLF模型的性能,在其中加入偏差,构造BENLF模型。在大型商业数据集上的实验表明,ENLF和BENLF模型的预测精度和模型稀疏性等性能有显著提高。In big data prediction,the SGD_LF model is a common method to predict missing data in high-dimensional sparse matrices. Since the SGD_LF model only constrains the objective function by the L 2 regularization term,which cannot regulate the distribution of latent factor. Thus,the model fails to be so accurate to describe the entity features in the target matrix well. In order to solve this problem,this paper proposed a FOBOS algorithm to construct an elastic network ENLF model,using L 1 and L 2 to limit the objective function. ENLF regulated the distribution of latent factors and improved the performance of the model. To further improve the performance of the ENLF model,we integrated the bias into the ENLF and constructd the BENLF model. The experiments on large industrial datasets show that the prediction accuracy and model sparsity of the ENLF and BENLF models are significantly improved.

关 键 词：高维稀疏矩阵 隐因子 大数据 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]