多源异构签到数据中改进RNN的POI推荐

POI recommendation using improved RNN in multi-source heterogeneous sign-in data

在基于位置的社交网络(LBSNs)中,现有的兴趣点(POI)推荐方法主要考虑地理位置和社交关系因素的影响,对签到行为的顺序和时间因素影响关注较少。针对该问题,提出一种改进循环神经网络(RNN)的POI推荐的方法。通过因子分解机对影响POI推荐因素的稀疏矩阵进行去稀疏化;通过提出的MMBE框架对多源异构签到数据整体建模,得到POI推荐的影响因子;将影响因子输入改进型RNN,计算出兴趣点预测值,将预测值最高的前K个兴趣点推荐给用户。实验结果表明,所提方法在精度、召回率、F1值方面优于其它3种较新的POI推荐方法。

Existing research mainly focuses on the influence of geographical and social relationships on POI recommendation and pays less attention to the order and time of check-in behavior.Aiming at this problem,considering the influence of geographical location,social relationship,time and order on the selection of points of interest,a method for improving the POI recommendation of RNN in multi-source heterogeneous sign-in data was proposed.The factorization machine was used to preprocess the sparse matrix that affected the POI recommendation factors.The overall multi-source heterogeneous sign-in data were modeled through the proposed MMBE framework,and the POI recommendation impact factors of all the above factors were integrated.The impact factor was substituted into the improved RNN to calculate the predicted value of interest points,and the top K inte-rest points with the highest predicted value were recommended to the user.Experimental results show that the proposed method is better than the other three newer POI recommendation methods in terms of accuracy,recall rate,and F 1 value.