ISSN 1000-1239 CN 11-1777/TP

计算机研究与发展 ›› 2018, Vol. 55 ›› Issue (2): 391-399.doi: 10.7544/issn1000-1239.2018.20160928

• 软件技术 • 上一篇    下一篇


王海艳1,2,3, 肖亦康1   

  1. 1(南京邮电大学计算机学院 南京 210023); 2(江苏省无线传感网高技术研究重点实验室 南京 210003); 3(江苏省大数据安全与智能处理重点实验室 南京 210023) (
  • 出版日期: 2018-02-01
  • 基金资助: 

Dynamic Group Discovery Based on Density Peaks Clustering

Wang Haiyan1,2,3, Xiao Yikang1   

  1. 1(School of Computer Science, Nanjing University of Posts and Telecommunications, Nanjing 210023); 2(Jiangsu High Technology Research Key Laboratory for Wireless Sensor Networks, Nanjing 210003); 3(Jiangsu Key Laboratory of Big Data Security & Intelligent Processing, Nanjing 210023)
  • Online: 2018-02-01

摘要: 近年来,群组推荐由于其良好的实用价值得到了广泛关注.群组发现作为群组推荐的前提环节,其发现结果对推荐效果有着至关重要的影响,群组相似度越高,推荐的效果和稳定性越好.针对现有群组发现方法中存在忽略用户倾向具有时间迁移性和群组可重叠性展开研究,提出了一种基于密度峰值聚类的动态群组发现方法.该方法首先通过动态泊松分解得到量化的用户动态倾向,然后通过高阶奇异值分解预测不同的时间节点下用户对不同项目的倾向,并根据计算所得的用户倾向构建高相似度用户集合,最后利用改进的基于密度峰值的聚类算法对用户集合进行划分,实现群组发现.仿真实验对比结果表明:上述基于密度峰值聚类的群组发现方法具有更好的群组推荐效果.

关键词: 时间上下文, 动态性, 相似度, 密度峰值聚类, 群组发现

Abstract: Group recommendation has recently received wide attention due to its significance in real applications. As a premier step of group recommendation, group discovery is very important and discovery results will impact a lot on the performance of group recommendation. The higher similarity the groups have, the better effectiveness and stability the recommendation results will possess. However, current group discovery methods seldom consider the dynamicity of users’ tendency with variance of time context, nor do they support the existence of groups overlapping. In order to address the problems above, a dynamic group discovery method based on density peaks clustering (DGD-BDPC) is put forward in this paper. In the proposed DGD-BDPC method, quantitative users’ dynamic tendency is firstly obtained by dynamic poisson factorization. And secondly, users’ tendency under different time nodes for various items will be predicted with the employment of high order singular value decomposition (HOSVD) and user sets with high similarity will then be built according to users’ tendency. Finally, user sets will be clustered with a modification of density peaks clustering algorithm and group discovery will be realized successfully. Experimental results show that the proposed dynamic group discovery method based on density peaks clustering has higher accuracy, lower error and better stability compared with some other methods.

Key words: time context, dynamicity, similarity, density peaks clustering, group discovery