基于神经网络和社区发现的高维数据推荐系统  

RECOMMENDATION SYSTEM OF HIGH DIMENSIONAL DATA BASED ON NEURAL NETWORKS AND COMMUNITY DISCOVERY

作　　者：唐新宇 张新政[2] 刘保利[3] Tang Xinyu;Zhang Xinzheng;Liu Baoli(School of Computer Engineering,Guangdong Business and Technology University,Zhaoqing 526040,Guangdong,China;School of Automation,Guangdong University of Technology,Guangzhou 510090,Guangdong,China;College of Science,Air Force Engineering University,Xi’an 710000,Shaanxi,China)

机构地区：[1]广东工商职业技术大学计算机工程学院,广东肇庆526040 [2]广东工业大学自动化学院,广东广州510090 [3]空军工程大学理学院,陕西西安710000

出　　处：《计算机应用与软件》2020年第7期232-239,共8页Computer Applications and Software

基　　金：广东省教育厅高校特色创新类项目(自然科学)(2017GKTSCX110);广东省省级科技计划项目(2014A020217016)。

摘　　要：由于评分矩阵存在稀疏性问题和冷启动问题,传统的推荐系统大多通过分析上下文环境来增强推荐系统的性能,导致计算复杂度提高,并影响推荐的准确率。针对这种情况,提出基于神经网络和社区发现的高维数据推荐系统。利用神经网络识别影响力大的上下文维度,提高预测的准确率;设计社区检测算法将用户分组,降低数据维度并解决稀疏性问题;采用张量模型处理包含丰富附加信息的用户评分矩阵,根据张量值预测用户对项目的偏好。仿真实验结果表明,该系统有效地提高了高维数据推荐系统的性能。Because of the existence of sparsity problem and cold start problem of rating matrix,most of classical recommendation systems enhance the performance by analyzing context environment,which leads to the increase of computational complexity and affects the accuracy of recommendation.In view of this,we propose a recommendation system of high dimensional data based on neural networks and community discovery.It used neural networks to recognize influential context dimensions,and improved the prediction accuracy;a community detection algorithm was designed to group users,and it could reduce data dimension and solve the sparsity problem;the tensor model was used to process the user rating matrix with additional information,and the user preference for the project was predicted according to the tensor value.The simulation results show that the system effectively improves the performance of high-dimensional data recommendation system.

关 键 词：推荐系统 稀疏性问题 社区发现 神经网络 张量模型 奇异值分解 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]