摘要
为了有效地捕捉LBSN中丰富的签到、社交等多维上下文信息的空间特性,并深层挖掘用户和POI之间的非线性交互,提出了一种谱嵌入增强的POI推荐算法——PSC-SMLP,设计了偏好增强的谱聚类算法PSC和谱嵌入增强的神经网络SMLP。在2个经典数据集上与现有的POI推荐算法相比,PSC-SMLP可以深层学习用户对POI的个性化偏好,在准确率、召回率、nDCG、平均精度等指标中均获得较大提升。
In order to effectively capture the spatial characteristics of multi-dimensional context information in LBSN,and deeply explore the non-linear interaction between users and POIs,a spectral embedding enhanced POI recommendation algorithm,namely PSC-SMLP,was proposed.A preference enhanced spectral clustering algorithm(PSC)and a novel spectral embedded enhanced neural network(SMLP)was designed to solve the above problems.Compared with state-of-the-art algorithms on two datasets,PSC-SMLP has better performance in terms of the precision,recall,nDCG and mean average precision.
作者
刘真
王娜娜
王晓东
孙永奇
LIU Zhen;WANG Na’na;WANG Xiaodong;SUN Yongqi(School of Computer and Information Technology,Beijing Jiaotong University,Beijing 100044,China)
出处
《通信学报》
EI
CSCD
北大核心
2020年第3期197-206,共10页
Journal on Communications
基金
国家重点研发计划基金资助项目(No.2019YFB2102501)。
关键词
基于位置的社交网络
POI推荐
谱聚类
谱嵌入
神经网络
location-based social network
POI recommendation
spectral clustering
spectral embedding
neural network