基于自适应高阶隐式关系建模的社交推荐系统(长文投稿)
作者:
作者单位:

1.安徽工程大学;2.同盾科技有限公司;3.华东师范大学

基金项目:

国家自然科学基金青年科学基金项目


Adaptive High-order Implicit Relations Modeling for Social Recommender System
Fund Project:

National Natural Science Foundation of China Youth Fund

  • 摘要
  • | |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • | |
  • 文章评论
    摘要:

    近年来,社交推荐的研究主要聚焦于社交网络中显式、隐式关系的联合建模,却忽视了高阶隐式关系并非对每个用户都同等重要这一特殊现象.高阶隐式关系对一个有着足够多邻居的用户与一个有着少量邻居的用户重要性存在明显差异.此外,由于社交关系建立的随机性,显式关系并不总是可用的.本文提出了一种新的自适应高阶隐式关系建模方法(Adaptive High-order Implicit Relations Modeling,简称 AHIRM),该模型由三个部分组成:首先,过滤不可靠关系且识别出潜在可靠关系.旨在避免不可靠关系带来的负面影响,并部分缓解数据稀疏的问题.其次,设计自适应随机游走算法,结合规范化后的节点中心度为用户捕获不同阶数的邻居,构建用户间的高阶隐式关系,进而重构社交网络.最后,运用图卷积网络(Graph Convolutional Network,简称 GCN)聚合邻居节点信息更新用户嵌入,实现高阶隐式关系建模,从而进一步缓解数据稀疏问题.在建模过程中,同时考虑到社交结构和个人偏好的影响,模拟并保留了社交影响传播的过程.本文在 LastFM,Douban 这两个数据集上与相关算法做了对比验证,结果证实了该模型的有效性和合理性.

    Abstract:

    Recently, the existing works have developed various social recommender systems that have primarily focused on explicit-implicit relations jointly modeling in general and overlooked the special phenomenon that high-order implicit relations are not equally important to each user. The importance of high-order implicit relations has a significant difference between the users with enough neighbors and the users with few neighbors. In addition, due to the randomness of explicit relations construction, which are not always available. In this paper, we propose a novel Adaptive High-order Implicit Relations Modeling (AHIRM) method. Our model consists of three components. First, by filtering unreliable relations and identifying potential reliable relations, we mitigate the adverse effects of unreliable relations and partially alleviate the data sparsity issue. Second, we design an adaptive random walk algorithm incorporating the normalized node centrality to capture the neighbors with different order for users adaptively, construct high-order implicit relations and reconstruct the social networks. Third, we employ GCN to integrate information from neighbors to update user embeddings modeling high-order implicit relations to further alleviate the data sparsity issue. Meanwhile, considering the influence of social structure and personal preference, we simulate and preserve the process of social influence propagation. Finally, we conduct extensive experiments on LastFM and Douban datasets. Both quantitative results and qualitative analysis verify the effectiveness and rationality of our AHIRM method.

    参考文献
    相似文献
    引证文献
引用本文
相关视频

分享
文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:2021-07-31
  • 最后修改日期:2021-11-30
  • 录用日期:2022-02-25
文章二维码
您是第位访问者
版权所有:中国科学院软件研究所 京ICP备05046678号-3
地址:北京市海淀区中关村南四街4号,邮政编码:100190
电话:010-62562563 传真:010-62562533 Email:jos@iscas.ac.cn
技术支持:北京勤云科技发展有限公司

京公网安备 11040202500063号