《计算机应用研究》|Application Research of Computers

基于多重降噪自编码器模型的top-<i>N</i>推荐算法

Top-<i>N</i> recommendation algorithm based on multiple denoising auto-encoder

免费全文下载 (已被下载 次)  
获取PDF全文
作者 方义秋,俞晨曦,葛君伟
机构 重庆邮电大学 计算机科学与技术学院,重庆 400065
统计 摘要被查看 次,已被下载
文章编号 1001-3695(2020)12-012-3582-04
DOI 10.19734/j.issn.1001-3695.2019.09.0534
摘要 针对传统降噪自编码器(DAE)中加噪操作的随机性影响预测精度的问题,以及数据矩阵忽视用户具体评分信息的问题,提出了一种结合用户评分的多重降噪自编码器(MDAE)。首先,在输入矩阵中引入具体评分信息,增加输入矩阵信息量;其次,为了在获得鲁棒性数据的前提下减轻加噪操作对预测精度的影响,构建了MDAE模型,将经过不同层次降噪得到的预测矩阵结合非降噪预测矩阵得出最终的预测结果;最后,将模型与其他模型在不同数据集上作实验对比。实验结果表明,结合用户具体评分的MDAE模型可以获得更优质的推荐结果。
关键词 预测精度; 用户评分; 加噪操作; 多重降噪自编码器
基金项目 重庆市基础与前沿研究计划资助项目(cstc2015jcyjA30001)
本文URL http://www.arocmag.com/article/01-2020-12-012.html
英文标题 Top-<i>N</i> recommendation algorithm based on multiple denoising auto-encoder
作者英文名 Fang Yiqiu, Yu Chenxi, Ge Junwei
机构英文名 College of Computer Science & Technology,Chongqing University of Posts & Telecommunications,Chongqing 400065,China
英文摘要 Focused on the problems that the randomness of the noise-adding operation in the traditional denoising auto-encoder(DAE) affects the prediction accuracy, and the data matrix ignores the user rating information, this paper proposed a multiple denoising auto-encoder combined with the user rating. Firstly, the method introduced the user rating information into the input matrix to increase the information quantity of the input matrix. Secondly, in order to reduce the corruption data effection on the prediction accuracy under the premise of obtaining robust data, it constructed the MDAE model. The prediction matrix with noisy data combined with the non-noise prediction matrix to obtain the final prediction result. Finally, the experiments compared the model with other models on different data sets. The experimental results show that the MDAE model combines with user rating information can get a better recommendation effect.
英文关键词 prediction accuracy; user rating; noise-adding operation; multiple denoising auto-encoder
参考文献 查看稿件参考文献
 
收稿日期 2019/9/9
修回日期 2019/11/5
页码 3582-3585
中图分类号 TP183
文献标志码 A