ImNext: Irregular Interval Attention and Multi-task Learning for Next POI Recommendation

作者全名：He, Xi; He, Weikang; Liu, Yilin; Lu, Xingyu; Xiao, Yunpeng; Liu, Yanbing

作者地址：[He, Xi; Lu, Xingyu; Liu, Yanbing] Chongqing Univ Posts & Telecommun, Sch Comp Sci & Technol, Chongqing 400065, Peoples R China; [He, Weikang; Xiao, Yunpeng] Chongqing Univ Posts & Telecommun, Sch Commun & Informat Engn, Chongqing 400065, Peoples R China; [Liu, Yilin] Beijing Univ Posts & Telecommun, State Key Lab Networking & Switching Technol, Beijing 100876, Peoples R China; [Liu, Yanbing] Chongqing Med Univ, Sch Med Informat, Chongqing 400016, Peoples R China

通信作者：Liu, YB (通讯作者)，Chongqing Univ Posts & Telecommun, Sch Comp Sci & Technol, Chongqing 400065, Peoples R China.; Liu, YB (通讯作者)，Chongqing Med Univ, Sch Med Informat, Chongqing 400016, Peoples R China.

来源：KNOWLEDGE-BASED SYSTEMS

ESI学科分类：COMPUTER SCIENCE

WOS号：WOS:001225514700001

JCR分区：Q1

影响因子：7.2

年份：2024

卷号：293

期号：　

开始页：　

结束页：　

文献类型：Article

关键词：Point-of-interest; Next POI recommendation; Irregular interval; Attention mechanism; Multi-task learning

摘要：The next point-of-interest (POI) recommendation task recommends users POIs that they may be interested in based on their historical trajectories. This task holds value for users as well as service providers; however, it is difficult. Although users exhibit repetitive and periodic behavioral characteristics at the macro level, such characteristics are influenced by individual preferences and diverse factors at the micro level, rendering prediction difficult. Most existing sequence modeling methods consider intervals between elements to be invariants. However, the time and distance intervals between adjacent POIs in the user's check-in sequences are irregular, which contains significant user behavioral characteristics. Therefore, we propose a model known as I rregular Interval Attention and M ulti-task Learning for Next POI Recommendation (ImNext). First, to address data sparsity and irregular intervals in the check-in sequence, we designed a data augmentation method to improve data density and proposed a novel irregular interval attention (IrrAttention) module. Second, to deal with the potential factors that affect user behavior, we proposed a graph attention network module that integrates edge attention (EA-GAT), which incorporates edge weights in the user's spatiotemporal and social transition graphs. Lastly, we established multiple subtasks for joint learning as the user's next check-in hides multiple targets, such as time and distance intervals. The experimental results show that our proposed method outperforms the state-of-the-art (SOTA) methods on two real-world public datasets. The implementation of the ImNext model is available at https://github.com/simplehx/ImNext.

基金机构：National Natural Science Foundation of China [62272074]; Chongqing Post-graduate Research and Innovation Project [CYB23235]

基金资助正文：<B>Acknowledgments</B> This paper is partially supported by the National Natural Science Foundation of China (Grant No. 62272074) and the Chongqing Post-graduate Research and Innovation Project (Grant No. CYB23235) .