摘要
协同过滤推荐算法对于含噪声稀疏数据集的推荐性能较弱,为此设计噪声检测修正和神经网络的稀疏数据top-k推荐算法。将用户和项目按评分分别分类为高分类、中等类和低分类,根据分类结果检测评分矩阵的奇异点,对奇异点做简单地修正处理。建立基于兴趣关系的受限玻尔兹曼机模型,将用户对项目的兴趣关系以及项目的次级信息作为条件受限玻尔兹曼机的输入,预测目标用户的top-k推荐列表。基于多个数据集的实验结果表明,该算法有效地提高稀疏数据的推荐性能,并且推荐列表的排序也较为准确。
Collaborative filtering recommendation algorithm has weak recommendation performance for noisy sparse datasets.Therefore,we design a top-k recommendation algorithm of sparse data based on the noise detection correction and neural networks.It classified users and items into high-rating class,middle-rating class and low-rating class,respectively,and detected the singularity points of rating matrices according to the classification results.Then,singularity points were simply modified.We established a restricted Boltzmann machine based on the interest relationships,and we set the interest relationships of users and the additional information of items as the inputs of the restricted Boltzmann machine.The top-k recommendation list of target users was predicted.Experimental results based on several datasets show that our algorithm improves the recommendation performance of sparse data effectively,and the rankings of recommendation lists are more accurate.
作者
张艳红
俞龙
Zhang Yanhong;Yu Long(School of Computer Science and Engineering,Tianhe College of Guangdong Polytechnic Normal University,Guangzhou 510540,Guangdong,China;College of Electronic Engineering,South China Agricultural University,Guangzhou 510642,Guangdong,China)
出处
《计算机应用与软件》
北大核心
2020年第8期274-281,共8页
Computer Applications and Software
基金
广东省科技计划项目(2013B020314014)
广东省教育科学规划教育信息技术研究专项(14JXN060)
广东技术师范大学天河学院计算机科学与技术重点学科建设项目(Xjt201702)。
关键词
协同过滤推荐系统
噪声数据集
稀疏数据集
噪声过滤
神经网络
受限玻尔兹曼机
Collaborative filtering recommendation system
Noisy dataset
Sparse dataset
Noise filtering
Neural networks
Restricted Boltzmann machine
