摘要
针对现有序列推荐研究中未充分考虑时间间隔信息和序列间项目交互关系的问题,提出一种融入时间间隔的跨序列推荐方法,该方法由个体序列、跨序列交互建模和线性融合3部分组成。在个体序列中,利用Transformer模型捕获项目特征和时间间隔信息,获取用户的个体偏好;在跨序列交互建模中,采用图神经网络和自注意机制捕获项目间的依赖关系,得到用户的全局偏好;通过线性融合个性和全局偏好预测用户的最终偏好。在4个公开数据集上的实验结果表明,该方法优于最佳基线,验证了其有效性。
Aiming at the problem that time interval information and inter-sequence item interactions are not fully considered in existing sequence recommendation studies,a cross-sequence recommendation method incorporating time intervals was proposed,which consisted of three parts of individual sequences,cross-sequence interaction modeling and linear fusion.In individual sequences,the Transformer model was used to capture item characteristics and time interval information to obtain users' indivi-dual preferences.In the cross-sequence interaction modeling,the graph neural network and self-attention mechanism were used to capture the dependencies between items to get the user's global preference.The user's final preference was predicted by linearly fusing individual and global preferences.Experimental results on four public datasets show that the method outperforms the optimal baseline,validating its effectiveness.
作者
贾丽云
佟玉军
李雪
吴金霞
周军
JIA Li-yun;TONG Yu-jun;LI Xue;WU Jin-xia;ZHOU Jun(School of Electronics and Information Engineering,Liaoning University of Technology,Jinzhou 121001,China;School of Science,Liaoning University of Technology,Jinzhou 121001,China)
出处
《计算机工程与设计》
北大核心
2025年第3期819-825,共7页
Computer Engineering and Design
基金
国家自然科学基金青年基金项目(61903167)。
关键词
序列推荐
时间间隔
跨序列交互
推荐系统
图神经网络
注意力机制
多层感知机
sequential recommendation
time interval
interaction across sequences
recommender systems
graph neural networks
attentional mechanisms
multilayer perceptual machines
