时空序列嵌入门控扩张残差网络的兴趣点推荐  

POI Recommendation of Spatiotemporal Sequence Embedding in Gated Dilation Residual Network

作　　者：姜涛 徐胜华[1] 李晓燕 张志然 王勇[1] 罗安[1] 何璇 JIANG Tao;XU Shenghua;LI Xiaoyan;ZHANG Zhiran;WANG Yong;LUO An;HE Xuan(Research Center of Geospatial Big Data Application,Chinese Academy of Surveying and Mapping,Beijing 100830,China;School of Resource and Environmental Sciences,Wuhan University,Wuhan 430079,China;School of Geomatics,Liaoning Technical University,Fuxin 123000,China;School of Earth Sciences and Engineering,Xi'an Shiyou University,Xi'an 710065,China)

机构地区：[1]中国测绘科学研究院地理空间大数据应用研究中心,北京100830 [2]武汉大学资源与环境科学学院,湖北武汉430079 [3]辽宁工程技术大学测绘与地理科学学院,辽宁阜新123000 [4]西安石油大学地球科学与工程学院,陕西西安710065

出　　处：《武汉大学学报（信息科学版）》2024年第9期1683-1692,共10页Geomatics and Information Science of Wuhan University

基　　金：国家自然科学基金(42071384)。

摘　　要：个性化兴趣点推荐是基于位置社交网络的一项重要服务,通过用户签到数据的序列信息和时空上下文信息可以有效挖掘用户的移动模式和兴趣偏好。为充分挖掘用户的个性化时空偏好和连续签到序列的长期依赖关系,提出嵌入时空条件的概率生成模型,构建门控扩张残差网络,实现基于门控扩张残差网络的兴趣点推荐方法。所提方法通过门控扩张残差网络学习用户的签到序列,将用户连续签到的空间距离和时间间隔作为约束条件,挖掘用户连续签到行为的时空规律,捕获用户签到行为的序列偏好和时空偏好。使用Foursquare和Instagram两套公开的签到数据集进行实验,结果表明,与表现最好的对比算法NextItNet相比,所提方法在召回率、精确度、F1分数和归一化折损累计增益等评价指标上都有明显提升。在Foursquare数据集上,各项指标的提升范围为1.52%~24.95%;在Instagram数据集上,各项指标的提升范围为7.06%~42.47%。所提方法适用于挖掘用户连续签到中存在的长期依赖关系,可以有效嵌入空间距离和时间间隔影响因素,提高了兴趣点推荐的准确性。Objectives:Personalized point of interest(POI)recommendation is a vital service in locationbased social network.It can effectively use the sequence and spatiotemporal context information of check-in data to discover movement patterns and preferences of users.Methods:This paper proposes a probabilistic generative model with embedded spatiotemporal conditions to fully exploit the long-term dependency between personalized spatiotemporal preferences and sequential check-in sequences of users,constructs a gated dilation residual network,and implements a POI recommendation method based on gated dilation residual network.The method in this paper learns check-in sequences of users through a gated dilation residual network.It mines and captures the spatiotemporal patterns,sequence preferences and temporal preferences constrained by the spatial distance and time interval of sequential check-in behavior of users.Results:The proposed method shows significant improvements on the Foursquare and Instagram datasets.Compared to the best-performing algorithm NextItNet,our method demonstrates noticeable enhancements in terms of recall,precision,F1 score,and normalized discounted cumulative gain.On the Foursquare dataset,we achieve improvements ranging from 1.52%to 24.95%.On the Instagram dataset,the improvements range from 7.06%to 42.47%.Conclusions:The proposed method is more suitable for mining the long-term dependency relationships in sequential check-in behavior of users.It effectively incorporates spatial distance and temporal interval factors,thereby improving the accuracy of POI recommendation.

关 键 词：兴趣点推荐 空间距离 时间间隔 门控扩张残差 时空序列 

分 类 号：P208[天文地球—地图制图学与地理信息工程]