LBSN中基于元路径的兴趣点推荐  

POI Recommendation Based on Meta-Path in LBSN

作　　者：曹玖新[1] 董羿[1] 杨鹏伟[1] 周涛[1] 刘波[1] 

机构地区：[1]东南大学计算机科学与工程学院计算机网络和信息集成教育部重点实验室(93K-9),南京211189

出　　处：《计算机学报》2016年第4期675-684,共10页Chinese Journal of Computers

基　　金：国家自然科学基金(61272531;61202449;61272054;61370207;61370208;61300024;61320106007;61472081);国家"九七三"重点基础研究发展规划项目基金(2010CB328104);国家"八六三"高技术研究发展计划项目基金(2013AA013503);高等学校博士点学科专项科研基金(2011009213002);江苏省科技计划项目基金(SBY2014021039-10);江苏省网络与信息安全重点实验室(BM2003201);计算机网络和信息集成教育部重点实验室(东南大学)(93K-9)资助

摘　　要：兴趣点(Point-Of-Interest,POI)推荐是基于位置的社交网络(Location-Based Social Networks,LBSN)中的一项重要个性化服务.由于LBSN中数据的极度稀疏性,基于协同过滤的算法推荐精度不高,文中提出基于元路径的兴趣点推荐算法.首先根据LBSN结构特征构建带权异构网络模型,其次引入元路径来描述节点间不同类型关联关系,基于三度影响力设置用户-兴趣点间元路径特征集,然后通过随机游走方法计算元路径特征值以度量实例路径中的首尾节点间关联度,并利用监督学习方法获得各特征的权值,最后计算特定用户将来在各兴趣点的签到概率从而生成推荐列表.文中在3个真实LBSN签到数据集上进行了实验,结果表明该算法可以有效缓解LBSN中的极度稀疏性问题,比传统推荐算法有更好的推荐效果.POIs recommendation is a crucial personalized service of Location-Based Social Networks（LBSN）.The extreme data sparsity in LBSN presents a big challenge to traditional collaborative filtering recommendation algorithms.To this end,we propose a novel POIs recommendation algorithm based on meta-path.Firstly,LBSN is considered as a weighted heterogeneous network according to its structure attributes,and the meta-path is introduced to describe different relations between nodes.Then the set of meta-paths between user and POI nodes is given on the basis of three degrees of influence.In addition,we calculate the eigenvalues of meta-path by random walk to measure the relevancy between the head and end nodes in a path instance,and determine the weight of each meta-path feature by supervised learning.At last,the check-in probability to each candidate POI is calculated to achieve the recommendation result.To evaluate our algorithm,we implement experiments on three real datasets.The results prove that our method can effectively alleviate the sparsity problem and improve the accuracy of recommendation system.

关 键 词：基于位置的社交网络 异构网络 兴趣点推荐 元路径 数据挖掘 社会媒体 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]