基于嵌入式向量和循环神经网络的用户行为预测方法  

User behavior prediction method based on embedded vector and recurrent neural network

作　　者：刘杨涛[1] 南书坡[2] 杨新锋[3] 

机构地区：[1]南阳理工学院软件学院,河南南阳473004 [2]河南师范大学新联学院公共教学部,河南郑州450000 [3]南阳理工学院计算机与信息工程学院,河南南阳473004

出　　处：《现代电子技术》2016年第23期165-169,共5页Modern Electronics Technique

基　　金：河南省科技攻关重点计划项目(122102210563;132102210215);河南省高等学校重点科研项目计划(15B520008)

摘　　要：为了进一步描述时间因素对用户行为的影响,进而提高推荐系统的推荐效果,综合考虑了用户的长期行为特征和短期行为特征提出一种基于嵌入式向量和循环神经网络的用户行为预测方法。依据推荐系统中的所有用户行为数据,将用户和商品嵌入到相同的特征空间,并通过嵌入式向量反应用户的长期行为特征。针对每个用户,依据其历史行为的时间序列,基于循环神经网络建立该用户的行为预测模型,从而描述该用户的短期行为特征。实验结果表明,提出的方法与特征级时间序列分析等方法相比具有更好的推荐效果。In order to further describe the effect of time factor on user behavior, and improve the recommendation effect of the recommendation system, a user behavior prediction method based on embedded vector and recurrent neural network is pro- posed by comprehensively considering the long-term and short-term user behavior features. According to all user behavior data in recommendation system, the users and commodities are embedded into the same feature space. The embedded vector is used to re- fleet the long-term user features. According to the time series of each user＇s historic behavior, the user behavior prediction model was built based on recurrent neural network to describe this user＇s short-term behavior feature. The experiment results show that, the proposed method has better recommendation effect than that of the feature-level time series analysis method and other methods.

关 键 词：循环神经网络 深度学习 嵌入式向量 用户行为预测 时间序列 

分 类 号：TN711-34[电子电信—电路与系统] TP319[自动化与计算机技术—计算机软件与理论]