本文已被：浏览 1103次 下载 327次
Received:March 18, 2020
Received:March 18, 2020
中文摘要: 随着互联网技术的飞速发展，互联网用户在畅游网络的同时也面临着信息过载的问题，而个性化推荐技术则成为了解决信息过载问题的有力工具。为了对用户提供更精准的商品推荐服务，提出了一个基于栈式降噪自编码器（SDAE）和贝叶斯个性化排序（BPR）相结合的深度神经网络模型SDAE-BPR。首先，使用SDAE把商品评分数据作为输入，编码后得到隐特征。其次，用BPR的方法学习对应商品的隐特征向量。该模型能够避免矩阵稀疏性的影响，因此达到了更精准推荐商品的效果。最后，分别使用Movielens 20 M和Movielens 1 M数据集，对SDAE-BPR模型与传统基于商品的协同过滤模型（IB-CF）、传统基于用户的协同过滤模型（UB-CF）做了对比，结果发现SDAE-BPR具有更高的准确度。
Abstract:With the development of Internet technology,the problem of information overload needs to be solved.Personalized recommendation technology has become a powerful tool to solve the problem of information overload.In order to provide more accurate commodity recommendation service for users,a deep neural network model (SDAE-BPR) based on the combination of stack denoising auto-encoder(SDAE) and Bayesian personalized ranking(BPR) is proposed.Firstly,SDAE is used to take the commodity score data as input,and the hidden features are obtained after coding.Secondly,the implicit feature vectors of the corresponding commodities are learned by the method of BPR.This model can avoid the influence of matrix sparsity,thus achieving the effect of more accurate commodity recommendation.Finally,based on Movielens 20 M data set,the results of SDAE-BPR are compared with the traditional item-based collaborative filtering model(IB-CF) and the traditional user-based collaborative filtering model(UB-CF).It can be seen that SDAE-BPR has higher accuracy.
keywords: recommendation system stack denoising auto-encoder Bayesian personalized ranking deep learning
文章编号：20215014 中图分类号：TP399 文献标志码：
基金项目:国家自然科学基金（61672337）。
Reference text: