ISSN 1000-1239 CN 11-1777/TP

计算机研究与发展 ›› 2016, Vol. 53 ›› Issue (4): 752-763.doi: 10.7544/issn1000-1239.2016.20151005

• 网络技术 • 上一篇    下一篇



  1. 1(武汉大学计算机学院 武汉 430072);2(计算机软件新技术国家重点实验室 (南京大学) 南京 210046);3(湖北大学教育学院 武汉 430062) (
  • 出版日期: 2016-04-01
  • 基金资助: 

A Synthetic Recommendation Model for Point-of-Interest on Location-Based Social Networks: Exploiting Contextual Information and Review

Gao Rong1, Li Jing1, Du Bo1, Yu Yonghong2, Song Chengfang1, Ding Yonggang1,3   

  1. 1Computer School, Wuhan University, Wuhan 430072); 2State Key Laboratory for Novel Software Technology (Nanjing University), Nanjing 210046); 3Faculty of Education, Hubei University, Wuhan 430062)
  • Online: 2016-04-01

摘要: 随着位置社交网络(location-based social network, LBSN)的快速增长,兴趣点(point-of-interest, POI)推荐已经成为一种帮助人们发现有趣位置的重要方式.现有的研究工作主要是利用用户签到的历史数据及其情景信息(如地理信息、社交关系)来提高推荐质量,而忽视了利用兴趣点相关的评论信息.但是,现实中用户在LBSN中只对少数兴趣点进行签到,使得用户签到历史数据及其情景信息极其稀疏,这对兴趣点推荐来说是一个巨大的挑战.为此,提出了一种新的兴趣点推荐模型,称为GeoSoRev模型.该模型在已有的基于矩阵分解的经典推荐模型的基础上,融合关于兴趣点的评论信息、用户社交关联和地理信息这3个因素进行兴趣点推荐.基于2个来自Foursquare的真实数据集的实验结果表明,与其他主流的兴趣点推荐模型相比,GeoSoRev模型在准确率和召回率等多项评价指标上都取得了显著的提高.

关键词: 地点推荐, 矩阵分解, 社交关系, 地理信息, 评论文本

Abstract: With the rapid growth of location-based social network (LBSN), point-of-interest (POI) recommendation has become an important mean to help people discover attractive locations. However, most of existing models of POI recommendation on LBSNs improve recommendation quality by exploiting the user check-in history behavior and contextual information(e.g., geographical information and social correlations), and they tend to ignore the review texts information accompanied with rating information for recommender models. While in reality, users only check in a few POIs in LBSN, which makes the user-POIs check-in history records and contextual information highly sparse, and causes a big challenge for POIs recommendations. To tackle this challenge, a novel POIs recommendation model called GeoSoRev is proposed in this paper, which combines users’ preference to a POI with geographical information, social correlations and reviews text on the basis of the classic recommendation model based on matrix factorization. Experimental results on two real-world datasets collected from Foursquare show that GeoSoRev achieves significantly superior precision and recalling rates compared with other state-of-the-art POIs recommendation models.

Key words: location recommendation, matrix factorization, social relationships, geographical information, review text