用于下一项推荐的序列感知深度网络  

Sequence-aware Deep Network for the Next-item Recommendation

作　　者：赵串串 游进国[1,2] 李晓武 ZHAO Chuan-chuan;YOU Jin-guo;LI Xiao-wu(Faculty of Information Engineering and Automation,Kunming University of Science and Technology,Kunming 650500,China;Computer Technology Application Key Lab of Yunnan Province,Kunming 650500,China)

机构地区：[1]昆明理工大学信息工程与自动化学院,昆明650500 [2]云南省计算机应用重点实验室,昆明650500

出　　处：《小型微型计算机系统》2020年第7期1389-1394,共6页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61462050,61562054)资助;云南省自然科学基金项目(KKSY201603016)资助。

摘　　要：推荐系统旨在解决项目的信息爆炸问题并为用户提供个性化推荐.通常,用户交互的历史项目对于用户交互的下一个项目是有着不同的影响的.为此,本文提出一种序列感知深度网络(SeqaDN).本算法主要分为三部分,即项目嵌入、序列感知和深度神经网络偏好学习.首先,通过Item2vec项目嵌入方法将项目的上下文项目作为输入,得到项目的嵌入;其次,本文在Seqa DN中添加了一个自注意力网络,以感知序列中每个项目关于当前下一个项目的不同影响权重;最后,基于深度双向循环神经网络学习用户的历史偏好.通过在真实数据集MovieLens上设置对比实验验证算法有效性,与传统算法以及同类型算法作对比,实验结果证明了SeqaDN比现有的序列推荐方法取得了更好的推荐性能.The recommendation system is designed to address the information explosion problem of projects and provide personalized recommendations for users.In general,the history items of user interaction have different effects on the next item of user interaction.To this end,this paper proposes a sequence-aware deep network(SeqaDN).The algorithm is mainly divided into three parts,namely project embedding,sequence sensing,and deep neural network preference learning.Firstly,the item embedding is obtained by taking the context item of the item as input through Item2 vec item embedding method.Secondly,this paper adds a self-attention network in SeqaDN to perceive different influence weights of each item in the sequence about the current next item.Finally,based on the Deep Bidirectional Recurrent Neural Network to learn the user’s historical preferences.By setting the contrast experiment on the real data set MovieLens to verify the effectiveness of the algorithm,compared with the traditional algorithm and the same type of algorithm,the experimental results show that SeqaDN has better recommendation performance than the existing sequence recommendation method.

关 键 词：下一项推荐 序列感知 自注意力网络 深度双向循环神经网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]