作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程

• 先进计算与数据处理 • 上一篇    下一篇

基于多特征分类的微博好友推荐

程倩倩,王路路,郑 涛,姬东鸿   

  1. (武汉大学计算机学院,武汉430072)
  • 收稿日期:2014-05-07 出版日期:2015-04-15 发布日期:2015-04-15
  • 作者简介:程倩倩(1989 - ),男,硕士,主研方向:个性化推荐,社交网络,数据挖掘;王路路、郑 涛,硕士;姬东鸿,教授、博士生导师。
  • 基金资助:
    国家自然科学基金资助重点项目“篇章级中文语义分析理论与方法”(61133012)。

Microblog Friend Recommendation Based on Multi-feature Classification

CHENG Qianqian,WANG Lulu,ZHENG Tao,JI Donghong   

  1. (Computer School,Wuhan University,Wuhan 430072,China)
  • Received:2014-05-07 Online:2015-04-15 Published:2015-04-15

摘要: 现有微博好友推荐算法使用的用户信息比较单一,不能充分利用微博用户信息来刻画用户特征,导致推荐效果不理想。为解决该问题,在综合分析用户标签信息、内容信息、交互信息以及社交拓扑信息的基础上,通过计算主题相关度、兴趣相关度、用户亲密度进行特征挖掘,并采用K 最近邻分类算法为目标用户进行微博好友推荐。在新浪微博真实用户数据集上的实验结果表明,该算法的准确率、召回率、F1 度量值分别为16. 5% ,26. 8% ,19. 2% ,推荐效果优于基于内容的推荐算法和基于社会过滤的推荐算法。

关键词: 多特征, 好友推荐, 主题相关度, 兴趣相关度, 亲密度, K 最近邻

Abstract: Existing microblog friend recommendation algorithms just use relatively simple information about user to make recommendation, they can not take full advantages of the rich user information and describes the profound characteristics of the user, and the result which in recommendation effect is not very satisfactory. Aiming at these problems,this paper presents an algorithm that uses the user’s tag,user’s content,the interaction between users and the user’s social topological information. It mines and computes topic relevancy,interest relevancy,intimacy score from these information and then uses K Nearest Neighbor(KNN) classification algorithm to recommend. Through experiments on Sina microblog user data set,and the results show that the algorithm can achieve better recommendation effect than content-based algorithm and social filtering-based algorithm,and the highest accuracy,recall,F1 measure respectively reach 16. 5% ,26. 8% ,19. 2% .

Key words: multi-feature, friend recommendation, topic relevancy, interest relevancy, intimacy, K Nearest Neighbor (KNN)

中图分类号: