摘要
随着电子商务的飞速发展,推荐算法成为推荐领域的研究热点之一,而冷启动问题与数据稀疏成为推荐算法面临的主要问题。有研究工作指出,利用矩阵分解来缓解评分数据的稀疏性,但仍存在数据量大且推荐精度不高的问题。因此,提出一种基于用户聚类的非负矩阵分解的推荐算法。在原有的非负矩阵分解模型上,它结合用户的评分数据,基于聚类思想对用户进行聚类,充分挖掘用户间的相关关系。所提算法在Movielens数据集上进行实验,结果表明,在数据稀疏的情况下,该算法在均方根误差(RMSE)评价指标上优于传统非负矩阵分解算法,且预测误差减少了4.5%,改善了推荐效果。
With the rapid development of e-commerce, the recommendation algorithm becomes one of the research hotspots in the recommendation field. Cold start and data sparseness are the main problems faced by the recommended algorithm. Some research work points out that the matrix decomposition may be used to alleviate the sparseness of scoring data, but still exist the problems of large data amount and low recommendation accuracy. Thus, a recommendation algorithm based on user clustering for non-negative matrix factorization is proposed. On the original non-negative matrix factorization model, this algorithm, by combining the user＇s scoring data and based on the idea of clustering, clusters the users and fully exploits the correlation between the users. Experiment on the Movielens data set indicates that this proposed algorithm is superior to the traditional non-negative matrix factorization algorithm in RMSE （Root Mean evaluation index when the data is sparse. The modified recommendation algorithm can reduce error by 4.5% and improves the recommendation effect. Square Error） the prediction
作者
骆孜
龙华
邵玉斌
杜庆治
LUO Zi;LONG Hua;SHAO Yu-bin;DU Qing-zhi(School of Information Engineering and Automation,Kunming University of Science and Technology,Kunming Yunnan 650500,China)
出处
《通信技术》
2018年第11期2675-2679,共5页
Communications Technology
基金
地区科学基金项目(No.61761025)~~
关键词
推荐系统
用户相似度
聚类
非负矩阵分解
recommendation system
user similarity
clustering
non-negative matrix factorization
