融合地理和时空信息的对比兴趣点推荐方法  

Contrastive point-of-interest recommendation incorporating geographic and spatio-temporal information

作　　者：闵昭浩 张䶮 MIN Zhao-hao;ZHANG Yan(School of Computer Science and Information Engineering,Hubei University,Wuhan 430062,China;Library of Hubei University,Hubei University,Wuhan 430062,China)

机构地区：[1]湖北大学计算机与信息工程学院,湖北武汉430062 [2]湖北大学图书馆,湖北武汉430062

出　　处：《计算机工程与设计》2025年第2期368-375,共8页Computer Engineering and Design

基　　金：国家自然科学基金项目(61977021);湖北省2019年技术创新专项基金项目(2019ACA144)。

摘　　要：针对兴趣点推荐中无法精准捕捉POI之间地理影响和高效学习用户-兴趣点(POI)交互行为动态表示的问题,提出一种融合地理和时空信息的对比兴趣点推荐方法(IGST-CL)。采用地理插值采样策略来缓解数据不平衡问题,利用一种动态消息传播机制的图卷积网络精准捕获地理影响。采用一种基于正余弦时间函数的时间注意力机制和超图网络联合学习用户-POI交互行为的动态表示。采用对比学习策略进一步增强模型性能。基于多任务学习方法自适应融合上述3个任务推断用户偏好。基于多个基准数据集的实验分析验证了IGST-CL模型相比其它主流兴趣点算法的优越性。Aiming at the problems of failing to accurately capture geographic influences among POIs and inefficiently learning dynamic representations of user-POI interaction behaviors in POI recommendation,a contrastive POI recommendation method that integrated geographic and spatio-temporal information(IGST-CL)was proposed.A geographic interpolation sampling stra-tegy was employed to alleviate the data imbalance problem,and a graph convolutional network with a dynamic message propagation mechanism was utilized to accurately capture geographic influences.A temporal attention mechanism based on the sin-cosin time function and a hypergraph network were employed to jointly learn a dynamic representation of user-POI interaction beha-vior.A contrastive learning strategy was employed to further enhance the model performance.User preferences were inferred based on a multi-task learning approach adaptively fusing the above three tasks.Experimental analyses based on two real-world datasets demonstrate the superiority of the IGST-CL model relative to other state-of-the-art POI recommendation algorithms.

关 键 词：兴趣点 超图卷积网络 时间注意力 消息传播 数据不平衡 正余弦函数 对比学习 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]