一种基于多信息融合的兴趣点推荐模型

Point of Interest Recommendation Model Based on Multi-information Fusion

兴趣点(POI)推荐为用户推荐未访问的POI,是基于位置的社交网络(LBSN)的基本问题.LBSN的迅速发展,使得大规模LBSN异构数据急剧增加,签到评分数据极其稀疏,如何充分利用LBSN中的异构数据,解决数据稀疏性的问题,提高推荐准确性是POI推荐的面临的挑战.本文首先对签到信息建立二分网络,学习用户和兴趣点的嵌入向量,得到用户对未访问兴趣点的评分,缓解签到数据的稀疏性.然后在签到数据中提取用户的签到序列,学习用户签到序列模式,进一步提高推荐准确性.最后利用Bayesian算法处理LBSN中的地理信息,并在前面的基础上建立统一的模型BiGloGeoRec融合这3种信息.本文在Weeplaces和Foursquare等不同数据集上的实验证明BiGloGeoRec模型的效果比其他POI推荐模型效果有较大提升.

Point-of-interest(POI)recommendation recommends unvisited POIs for users,which is the basic problem of location-based social networking(LBSN).With the rapid development of LBSN,modeling the joint effects of Heterogeneous information faces a severe challenge from the extreme sparsity of users'check-in data.In this paper,we propose a fused recommendation model termed BiGloGeoRec which fuse the access record,sequential and geographical aspects in a single model.In BiGloGeoRec,firstly,we build a bipartite Network with check-in data to learn the embedding vectors of users and POIs,so as to alleviate the sparsity of the sign in data.Then,we extracting POI sequence from check-in data and learning user's mobile mode to further improve the accuracy of recommendation.Finally,we deal with the geographical information with the Bayesian algorithm,and build a united POI recommendation model based on the previous.The experiments in this article on different data sets such as Weeplaces and Foursquare prove that the effect of the BiGloGeoRec model is much better than that of other POI recommendation models.