• 中国计算机学会会刊
  • 中国科技核心期刊
  • 中文核心期刊

计算机工程与科学

• 论文 • 上一篇    下一篇

基于上下文感知和个性化度量嵌入的下一个兴趣点推荐

鲜学丰1,陈晓杰2,赵朋朋2,杨元峰1 ,Victor S.Sheng3   

  1. (1.江苏省现代企业信息化应用支撑软件工程技术研发中心,江苏 苏州 215104;
    2.苏州大学智能信息处理及应用研究所,江苏 苏州 215006;
    3.阿肯色中央大学计算机科学系,康威 72035)
     
  • 收稿日期:2017-10-28 修回日期:2017-12-21 出版日期:2018-04-25 发布日期:2018-04-25
  • 基金资助:

    国家自然科学基金(61728205,61472268,61672372,61472211);江苏省高校自然科学面上项目(17KJD520009);苏州市产业技术创新专项(SYG201710)

Context-aware personalized metric
embedding for next POI recommendation

XIAN Xuefeng1,CHEN Xiaojie2,ZHAO Pengpeng2,YANG Yuanfeng1,Victor S.Sheng3   

  1. (1.Jiangsu Province Support Software Engineering R&D Center for
    Modern Information Technology Application in Enterprise,Suzhou 215104;
    2.Institute of Intelligent Information Processing and Application,Soochow University,Suzhou 215006,China;
    3.Department of Computer Science,University of Central Arkansas,Conway,AR 72035,USA)
  • Received:2017-10-28 Revised:2017-12-21 Online:2018-04-25 Published:2018-04-25

摘要:

随着基于位置的社交网络推荐系统的逐步发展,兴趣点推荐成为了研究热门。兴趣点推荐的研究旨在为用户推荐兴趣点,并且为商家提供广告投放和潜在客户发掘等服务。由于用户签到行为的数据具有高稀疏性,为兴趣点推荐带来很大的挑战。许多研究工作结合地理影响、时间效应、社会相关性等方面的因素来提高兴趣点推荐的性能。然而,在大多数兴趣点推荐的工作中,用户访问的周期性习惯和伴随用户偏好的上下文情境信息没有被深度地挖掘。而且,下一个兴趣点推荐中一直存在着数据的高稀疏度。基于以上考虑,针对用户签到的数据稀疏性问题,将用户周期性行为模式归结为上下文情境信息,提出了一种基于上下文感知的个性化度量嵌入推荐算法,同时将用户签到的上下文情境信息考虑进来,从而丰富有效数据,缓解数据稀疏性问题,提高推荐的准确率,并且进一步优化算法,降低时间复杂度。在两个真实数据集上的实验分析表明,本文提出的算法具有更好的推荐效果。
 

关键词: 基于位置的社交网络, 下一个兴趣点推荐, 推荐系统, 上下文感知, 度量嵌入

Abstract:

With the rapid development of LocationBased Social Networks (LBSN) recommender system, PointofInterest (POI) recommendation has become a hot topic. The research of POI recommendation aims to recommend POIs for users and to provide services such as advertising and potential customer discovery. Due to the high data sparseness of users’ checkins, POI recommendation faces a great challenge. Many researches combine geographical influence, time awareness, social relevance and other factors to improve the performance of POI recommendation. However, in most POI recommendation researches, the periodicity of mobility and the user preference varying with the change of contextual scenario have not been excavated in depth. Moreover, there exists high data sparseness in Next POI recommendation. Based on the above considerations, this paper proposes a Contextaware Personalized Metric Embedding (CPME) algorithm, which is based on the user's periodic behavior pattern. It takes into account the contextual information of users’ checkins, which can enrich the valid data, alleviate the data sparseness, improve the recommendation accuracy, and further optimize the algorithm to reduce the time complexity. The experimental analysis on two real LBSN datasets show that the proposed algorithm has better recommendation performance.

 

Key words: location-based social networks, next POI recommendation, recommender system, contextaware, metric embedding