基于IMD与Canopy的非均匀资源推荐算法  

Non-Uniform Resource Recommendation Algorithm Based on Improved Matrix Factorization and Canopy

作　　者：孙浩 周力青 SUN Hao;ZHOU Li-qing(College of Information Science and Engineering,Guilin University of Technology,Guilin Guangxi 541004,China)

机构地区：[1]桂林理工大学信息科学与工程学院,广西桂林541004

出　　处：《计算机仿真》2023年第10期477-481,共5页Computer Simulation

基　　金：2022年国家社科基金一般项目(22BTQ081)。

摘　　要：在向用户推荐资源的过程中,常见的推荐方法在不同相似性度量指标下,对于非均匀性资源的推荐结果质量不能得到保证。针对上述问题,提出基于改进矩阵分解与Canopy的非均匀资源推荐算法。利用Canopy技术对非均匀资源进行聚类处理,设定两个阈值作为判断依据,确定聚类中心,在所有聚类结果完全收敛后,获得非均匀资源的聚类结果。从聚类结果中提取出非均匀资源隐含特征,由改进矩阵分解技术将目标分解为两个子矩阵,并根据推荐需求定义损失函数,不断优化两个子矩阵,求解出最优结果,组成推荐集合,将结果推荐给用户。实验结果表明,设计的推荐算法在不同相似性度量指标下,top推荐预测效果好,更加接近用户需求,计算效率高,推荐质量得到了提高。In the process of recommending resources to users,traditional methods can't guarantee the quality of recommendation.Therefore,an algorithm of non-uniform resource recommendation based on improved matrix factorization and Canopy was put forward.First of all,Canopy technology was used to cluster non-uniform resources,and then two thresholds were set as the judgment basis for determining the cluster center.After all clustering results were completely converged,the clustering result of non-uniform resources was obtained.Then,hidden features of non-uniform resources were extracted from the result.On this basis,the target was decomposed into two sub-matrices by the improved matrix decomposition technology.Furthermore,the loss function was defined through recommendation requirements.Continuously,the two sub-matrices were optimized until optimal results were obtained.These results form a recommendation set.Finally,the result was recommended to users.Experimental results show that the designed algorithm has good prediction effect of top recommendation under different similarity metrics,which is nearer to the user's needs.Meanwhile,this algorithm has high calculation efficiency and high recommendation quality.

关 键 词：改进矩阵分解 非均匀性 推荐算法 数据处理 

分 类 号：TP274[自动化与计算机技术—检测技术与自动化装置]