一种预填补社团聚类的兴趣点推荐算法

Prefilling Community Clustering POI Recommendation Algorithm

基于位置的社交网络(LBSN)中的数据信息往往会存在数据稀疏,甚至部分信息缺失的情况,导致推荐的准确性不高.兴趣点推荐系统中蕴含着丰富的多源异构数据,如好友关系数据、地理位置数据以及用户对兴趣点的评分等,使用这些数据可以有效提升兴趣点推荐算法的准确率.本文提出一种预填补社团聚类的兴趣点推荐算法.通过社团聚类算法来分别对签到评分数据以及好友关系数据建模得到用户对兴趣点的个人兴趣和社交兴趣,并添加距离影响因素.从而建立了SoGS模型来进行兴趣点推荐.并且提出一种基于相容类的预填补算法来缓解原始用户-兴趣点评分矩阵的稀疏性问题,融合SoGS模型进行对比实验.实验采用Yelp数据集,结果表明,SoGS模型能有效提高兴趣点推荐系统的准确率和召回率.

The location-based social networks( LBSN) often have sparse data,even with some information missing,resulting in lowaccuracy of the recommendation. Points of interest( POI) recommendation system contains a wealth of heterogeneous data,such as friends relationship data,geographic data and POI’s score,making use of these data can effectively improve the accuracy of the POI recommendation algorithm. this paper proposes a prefilling and community clustering based POI recommendation approach. Through community clustering algorithm,we build user interest data and sign-in score data to get users’ interest and social interest in POIs,and add distance influence factors. we give a joint model SoGS( Social,Geographical and Score) to make the recommendation by fusing the data of social relationships,geographic location data and user-poi-rating matrix. Moreover,a prefilling algorithm based on compatible classes( PACC) is proposed to alleviate the sparsity problem of the original user-poi-rating matrix,and then fuse the SoGS model for comparative experiments. A contrast experiment is performed on the Yelp dataset. The result shows that,our model which adds PACC can improve the precision and recall rate.