融合社交关系与代价敏感的兴趣点推荐算法

POI recommendation algorithm with fusing social relation and cost-sensitive learning

在基于位置的社交网络中,用户签到矩阵极其稀疏,采用排序学习技术进行兴趣点推荐是目前的热门研究方向。针对基于排序学习的兴趣点推荐模型存在精度不高、推荐列表忽略兴趣点的位置等问题,提出一种基于ListMLE的兴趣点推荐算法。基于推荐列表中兴趣点位置的关注度差异,将改进ListMLE算法应用到兴趣点推荐中;用户社交关系影响融入ListMLE的打分函数;代价敏感方法融入推荐列表计算过程。实验表明,在真实数据集Gowalla上,算法的准确率和召回率均优于基线排序学习算法。

In location-based social networks,the check-in matrix of users is extremely sparse,so it is a hot research direction to recommend points of interest using sorting learning technology.However,the existing interest point recommendation model based on learning to rank has some problems,such as low accuracy,ignoring the position of interest points in the recommendation list and so on.In this paper,we propose a model of interest point recommendation based on ListMLE.First of all,considering the attention degree of interest points in the recommendation list,the improved ListMLE algorithm is applied to the recommendation of interest points.Secondly,the social relationship impact of users is integrated into the scoring function of ListMLE.Finally,the cost sensitive method is integrated into the recommendation list calculation process.Experiments show that the algorithm is better than the baseline sorting learning algorithm in accuracy and recall on real data sets Gowalla.