LBSN中利用深度学习的POI推荐方法

POI recommendation method using deep learning in LBSN

提出一种基于位置的社交网络中利用深度学习的POI推荐方法。建立一个地理时空注意力网络,以发现总体序列依赖性和微妙的POI-POI关系;将签到序列中连续的地理距离和时间间隔信息加入到地理时空注意力网络中,建立用户个性化移动行为和挖掘用户个性化时空偏好;设计特定于上下文的共同注意力网络,通过从签到历史中自适应选择相关签到活动来学习更改用户偏好,使地理-时空门控循环单元网络(geographical-spatiotemporal gated recurrent unit network,GS-GRUN)能够区分不同签到的用户偏好程度。在Foursquare和Gowalla数据集上的实验结果表明,所提算法能够显著提升POI推荐方法的推荐匹配度。

A POI recommendation method using deep learning in location-based social networks was proposed.A geo spatiotemporal attention network was established to simultaneously discover the overall sequence dependence and subtle POI-POI relationship.The continuous geographic distance and time interval information in the check-in sequence were added to GRU network to model user’s personalized mobile behavior and mine user’s personalized spatio-temporal preference.A context specific common attention network was designed to learn to change user preferences by adaptively selecting relevant check-in activities from the check-in history,which enabled GS-GRUN to distinguish different levels of user preferences.Experimental results on Foursquare and Gowalla datasets show that the proposed method can significantly improve the recommendation matching degree of POI recommendation method.