用户偏好与时间序列的兴趣点推荐模型  

User Preference and Time Sequence Based POI Recommendation Model

作　　者：陶丹[1] 姚伊 吴谨汐 范睿明 郑晨旺 TAO Dan;YAO Yi;WU Jin-xi;FAN Rui-ming;ZHENG Chen-wang(School of Electronic and Information Engineering,Beijing Jiaotong University,Beijing 100044,China)

机构地区：[1]北京交通大学电子信息工程学院,北京100044

出　　处：《小型微型计算机系统》2022年第3期582-588,共7页Journal of Chinese Computer Systems

基　　金：国家自然科学基金“面上”基金项目(61872027)资助。

摘　　要：传统的兴趣点推荐通常忽略了用户签到行为中序列模式的重要性,且无法有效地捕捉用户复杂且动态变化的兴趣偏好.由此,本文提出了一种用户偏好和时间序列的兴趣点推荐模型(User Preference&Time Sequence based POI Recommendation,UPTS-PRec).该模型能够分别对短期偏好和长期偏好建模并融合,以捕捉用户兴趣的变化.对于短期偏好,提出了融合时空上下文信息的长短期记忆网络来学习用户签到行为中复杂的序列转移模式,并通过基于目标的注意力机制进一步精确地提取短期偏好.对于长期偏好,基于用户注意力机制以捕捉用户和兴趣点之间细粒度的关系.最后,在Foursquare和Gowalla两个数据集上进行实验仿真.结果表明本文提出的UPTS-PRec模型和主流的推荐方法相比在不同的评价标准上性能有较好的提升,验证了所提出模型的有效性.Traditional Point-of-Interest(POI)recommendations usually ignore the importance of sequential patterns in user check-in behavior and fail to effectively capture the complex and dynamic interest preferences of the users.Motivated by this,this paper proposes a User Preference&Time Sequence based POI Recommendation model(UPTS-PRec),which can model and integrate long-term preferences and short-term preferences so as to capture the changes of users′interests.For short-term preferences,we propose a Long Short-Term Memory with spatio-temporal contextual information to learn complex sequential shift patterns in user check-in behaviors,and further accurately extract short-term preferences through the goal-based attention mechanism.For long-term preferences,a user attention based mechanism is used to capture the fine-grained relationship between users and POI.Finally,a series of experiments are carried out on Foursquare as well as Gowalla datasets.The results show that the performance of our proposed UPTS-PRec model outperforms those of other popular POI recommendation ones on different evaluation criteria,which testify the effectiveness of our model.

关 键 词：兴趣点推荐 循环神经网络 注意力机制 用户偏好 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]