基于序列图时空增强与地理关系的兴趣点推荐  

Points-of-interest recommendation based on spatial-temporal enhancement of sequence graph and geographical relationships

作　　者：刘超[1] 朱军 Liu Chao;Zhu Jun(College of Computer Science&Engineering,Chongqing University of Technology,Chongqing 400054,China)

机构地区：[1]重庆理工大学计算机科学与工程学院,重庆400054

出　　处：《计算机应用研究》2025年第3期755-761,共7页Application Research of Computers

基　　金：2021年重庆市社会科学规划一般项目(2021NDYB101)。

摘　　要：针对现有兴趣点(points-of-interest,POI)推荐存在的地理特征挖掘不充分与未将顺序信息纳入空间偏好的问题,提出基于序列图时空增强与地理关系(spatial-temporal enhancement of sequence graph and geographical relationships,STESGGR)的POI推荐模型。首先,利用POI位置信息构建地理图,采用图卷积网络(graph convolutional network,GCN)与注意力机制获取用户访问POI的地理特征。其次,利用用户签到信息提取时空特征构建时空信息增强的序列图,采用门图神经网络(gated graph neural network,GGNN)与注意力机制获取用户访问POI的时空偏好。然后,引入共同性学习优化框架学习顺序信息与地理特征之间的互补信息,进一步挖掘地理特征。最后,融合两个特征信息并通过多层感知机(multilayer perceptron,MLP)进行POI推荐。在五个真实数据集上进行了实验,结果表明STESGGR模型在AUC和Logloss指标上分别提升1.2%~2.7%和3.2%~12.4%。实验证明STESGGR在基于位置的POI推荐下有较好的表现,充分挖掘了顺序与地理特征,提升了推荐效果。To address the insufficient of mining geographical feature and the absence of incorporating sequential information into spatial preferences in existing POI recommendation methods,this paper proposed a POI recommendation model based on STESGGR.Firstly,the model constructed a geographic graph using the location information of POI and employed GCN and attention mechanisms to capture the geographical features of user visits to POI.Secondly,it extracted spatio-temporal features from user check-in information to construct a sequence graph enhanced with spatio-temporal information and applied GGNN and with attention mechanisms to capture the spatio-temporal preferences of users visiting POI.Then,it introduced a common learning optimization framework to learn the complementary information between sequential information and geographical features,further mining geographical characteristics.Finally,it fused the two types of feature information for POI recommendations through MLP.Experiments on five real-world datasets demonstrate that the STESGGR model improves by 1.2%~2.7%in AUC and 3.2%~12.4%in Logloss metrics.The results validate that STESGGR performs well in location-based POI recommendations,effectively mining sequential and geographical features and enhancing the recommendation performance.

关 键 词：POI推荐 时空信息 地理信息 顺序信息 注意力机制 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]