计算机工程 ›› 2018, Vol. 44 ›› Issue (11): 215-221.doi: 10.19678/j.issn.1000-3428.0048755

• 人工智能及识别技术 • 上一篇    下一篇

基于遍历约束与交互信息增强的社交网络表征算法

石立鹏a,王莉b   

  1. 太原理工大学 a.信息与计算机学院; b.大数据学院,山西 晋中 030600
  • 收稿日期:2017-09-21 出版日期:2018-11-15 发布日期:2018-11-15
  • 作者简介:石立鹏(1990—),男,硕士研究生,主研方向为社交网络、数据挖掘;王莉,教授、博士。
  • 基金项目:

    国家高技术研究发展计划(2014AA015204);山西省自然科学基金(2014011022-1)。

An Enhanced Social Network Representation Algorithm Based on Traversal Constraint and Interactive Information

SHI Lipenga,WANG Lib   

  1. a.College of Information and Computer; b.College of Data Science,Taiyuan University of Technology,Jinzhong,Shanxi 030600,China
  • Received:2017-09-21 Online:2018-11-15 Published:2018-11-15

摘要:

传统网络表征方法将网络的拓扑结构转换为邻接矩阵以进行特征提取,在准确率和效率上存在局限性。为此,通过分析自然语言处理模型word2vec和多种网络表征算法,结合社交网络的特征,提出一种基于遍历约束和交互信息的社交网络表征算法。对社交网络遍历进行约束以提高算法的时间效率,利用用户交互信息修改word2vec模型,提高社交网络表征的准确率。在BlogCatalog和新浪微博2个社交网络数据集上进行的实验结果表明,相对DeepWalk、Line算法,该算法在时间效率上提高约20%,在准确率上提高约12%。

关键词: 特征学习, 网络遍历, 自然语言处理, 交互信息, 社交网络, 网络表征

Abstract:

Traditional network representation methods transform network topology into adjacency matrix for feature extraction,which has limitations in accuracy and efficiency.Therefore,by analyzing the natural language processing model word2vec and a variety of network representation algorithms,combined with the characteristics of social networks,a social network representation algorithm based on traversal constraint and interactive information is proposed.Constraints are applied to the traversal of social networks to improve the time efficiency of the algorithm.word2vec model is modified by user interaction information to improve the accuracy of social network representation.Experimental results on two social network datasets,BlogCatalog and Sina Weibo,show that compared with DeepWalk and Line algorithm,this algorithm improves the time efficiency by about 20% and the accuracy by about 12%.

Key words: feature learning, network traversal, natural language processing, interactive information, social network, network representation

中图分类号: