基于时空循环神经网络的下一个兴趣点推荐方法

A recurrent neural network model based on spatial and temporal information for the next point of interest recommendation

下一个兴趣点推荐已经成为基于位置的社交网络(location-based social networks,LBSNs)中一个重要任务。现有的模型没有深入考虑相邻签到兴趣点之间的转移时空信息,无法对用户访问下一个兴趣点的长短时间偏好和远近距离偏好进行有效建模。本文通过对循环神经网络(recurrent neural network,RNN)进行扩展,提出一个新的基于会话的时空循环神经网络模型(sesson-based spatial-temporal recurrent neural network,SSTRNN)用于下一个兴趣点推荐。该模型通过设置时间转移矩阵和空间转移矩阵分别对用户的时间和空间偏好信息进行建模,综合考虑连续签到兴趣点的序列信息、时空信息以及用户偏好进行下一个兴趣点推荐。通过在2个真实公开的数据集上进行实验,结果显示本文提出的SST-RNN模型的推荐效果比主流的推荐模型有显著提升。在Foursquare和CA数据集上,ACC@5评价指标分别提升了36.38%和13.81%,MAP评价指标分别提升了30.72%和17.26%。

The next point-of-interest(POI)recommendation has become an important task in location-based social networks.The existing models lack in-depth research on the temporal and spatial information transition between adjacent check-in POIs and cannot effectively model the long/short time and distance preferences of the users accessing the next POI.In response,this paper proposes a new session-based spatial–temporal recurrent neural network(SST-RNN)model that is used to recommend the next POI.This model takes advantage of the spatial transition matrix and temporal transition matrix to respectively model the user’s spatial and temporal preferences,and comprehensively considers the sequence information and spatial–temporal information of consecutive check-in POIs as well as user preferences to do the next POI recommendation.Experimental results in two real open datasets show that the performance of the proposed SST-RNN model is significantly enhanced compared with the state-of-the-art models.On the Foursquare and CA datasets,the ACC@5 is increased by 36.38%and 13.81%,and the MAP is increased by 30.72%and 17.26%,respectively.