一种基于矩阵分解和随机森林算法的推荐模型  

A RECOMMENDATION MODEL BASED ON MATRIX DECOMPOSITION AND RANDOM FOREST ALGORITHM

作　　者：罗云芳[1] 李力[2] Luo Yunfang;Li Li(School of Mechanical Electronic&Information Engineering,Guangxi Vocational&Technical College,Nanning 530226,Guangxi,China;School of Computer Science and Engineering,University of Electronic Science and Technology of China,Chengdu 610054,Sichuan,China)

机构地区：[1]广西职业技术学院机电与信息工程学院,广西南宁530226 [2]电子科技大学计算机科学与工程学院,四川成都610054

出　　处：《计算机应用与软件》2021年第8期315-321,共7页Computer Applications and Software

基　　金：广西教育厅自然科学基金项目(桂教科研[2018]2号2018KY0951,桂教科研[2019]1号2019KY1220)。

摘　　要：针对传统环境下运行的推荐算法预测精度不高的问题,提出一种基于矩阵分解和随机森林算法的推荐模型。提出的基于数据分割策略和新的学习过程的分布式推荐模型是在Apache Spark上设计的。通过数据分区、模型训练和偏好预测三个步骤处理大规模数据,提高预测质量,解决数据稀疏问题。为了提高模型在大数据环境下的性能,采用基于矩阵分解(Matrix Factorization, MF)和随机森林(Random Forests, RF)混合的新颖学习过程,从而显著加快分布式训练的速度。实验结果表明,相对于其他算法,该算法在性能指标上具有明显的优势。Aiming at the problem that the prediction accuracy of the recommended algorithm running in the traditional environment is not high,a recommendation model based on matrix decomposition and random forest algorithm is proposed.The proposed distributed recommendation model based on data segmentation strategy and new learning process was designed on Apache Spark.It processed large-scale data through three steps:data partitioning,model training and preference prediction,which improved prediction quality and solved data sparseness.In order to improve the performance of the model in the big data environment,we used a novel learning process based on matrix factorization(MF)and random forests(RF),which significantly sped up the distributed training.The experimental results show that compared with other algorithms,the proposed algorithm has obvious advantages in performance indicators.

关 键 词：大数据 推荐系统 矩阵分解 随机森林 分布式计算 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]