摘要
目前电商数据存在维度多、实时性要求强等特点,很多传统的推荐算法并不能很好地适用于电商推荐。针对电商场景中需要同时考虑用户长期偏好和短期偏好,数据维度高导致推荐算法运行效率低,少数无关数据影响对用户真实意图的判断等问题,论文提出了一种基于GRU网络的会话型混合推荐算法。该混合推荐算法同时考虑了用户的长短期偏好,能够通过注意力机制推测用户真实意图,相比于其他循环神经网络推荐算法提高了运行效率,提高了推荐准确度。
At present,e-commerce data has many characteristics,such as multi-dimensional,real-time requirements and so on.Many traditional recommendation algorithms are not suitable for e-commerce recommendation.In order to solve the problems of users'long-term and short-term preferences,low efficiency of recommendation algorithm due to high data dimension,and the influence of a few irrelevant data on users'real intention judgment,a hybrid recommendation algorithm based on GRU network is proposed.The hybrid recommendation algorithm takes into account the users'long-term and short-term preferences,infers users'real intentions through attention mechanism,and improves the efficiency and accuracy of recommendation compared with other recurrent neural network recommendation algorithms.
作者
李镇宇
朱小龙
周从华
刘志锋
LI Zhenyu;ZHU Xiaolong;ZHOU Conghua;LIU Zhifeng(School of Computer Science and Telecommunication Engineering,Jiangsu University,Zhenjiang 212013;Jingkou New Generation Information Technology Industry Research Institute,Jiangsu University,Zhenjiang 212013;不详)
出处
《计算机与数字工程》
2022年第5期942-947,共6页
Computer & Digital Engineering
关键词
电商推荐
门限循环单元
注意力机制
长短期状态
e-commence recommendation
gated recurrent unit
attention mechanism
long-term and short-term status
作者简介
李镇宇,男,硕士,研究方向:推荐算法和大数据;朱小龙,男,硕士,副教授,研究方向:编码理论、软件系统构架和大型信息系统;周从华,男,博士,教授,研究方向:大数据技术、人工智能;刘志锋,男,博士,副教授,研究方向:模型检测、可信计算与物联网。
