计算机工程与应用2024，Vol.60Issue(20)：142-152,11.DOI:10.3778/j.issn.1002-8331.2307-0012
时间感知增强的动态图神经网络序列推荐算法
Time-Aware Enhancement Dynamic Graph Neural Networks for Sequential Recommendation Algorithm
摘要
Abstract
The existing graph neural network sequence recommendation methods only focus on the temporal information of sequences without considering the time interval information of sequences,and only use single-task mode for sequence recommendation,ignoring the auxiliary tasks that can enhance the data and improve the generalization ability,which will lead to the dynamic interaction preferences of users and interaction time information cannot be captured clearly.To allevi-ate the impact of these problems,a dynamic graph neural network sequence recommendation algorithm with enhanced time awareness(TaDGSR)is proposed,which has two advantages.Firstly,the method constructs a sequence as a dynamic graph with temporal information,incorporates time interval information and adds a time-gated attention network module to enhance the utilization of temporal information while capturing the higher-order dynamic connections between sequences.Secondly,the method adopts the long short-term prediction task with temporal threshold segmentation as its auxiliary task to enhance the utilization of temporal interval information representation,so that the model can better capture the dynamic preferences of users at different interaction intervals,and finally improve the performance of the sequence recommenda-tion task.Experiments on the Beauty dataset,Games dataset,and CDs dataset of Amazon.com show that:(1)the proposed method achieves average improvements of 4.47%,4.37%,2.72%,1.16%,4.61%and 3.97%in the Hit@10 and NDCG@10 metric values for the three datasets,respectively,compared to the current newer benchmark method;(2)adding time interval information can effectively improve the recommendation performance of dynamic graph neural networks,and the use of a time-gated attention module and a long-and short-term prediction assistance task can both bring positive improvements to the prediction performance.关键词
序列推荐/图神经网络/邻域聚合/时间感知/多任务学习Key words
sequential recommendation/graph neural network/neighborhood aggregation/time-aware/multi-tasking learning分类
信息技术与安全科学引用本文复制引用
陈万志,王军..时间感知增强的动态图神经网络序列推荐算法[J].计算机工程与应用,2024,60(20):142-152,11.基金项目
辽宁省教育厅高等学校基本科研项目(LJKZ0327). （LJKZ0327）
