[1]郭蕾蕾,俞 璐,段国仑,等.基于 AP 聚类的多特征融合方法[J].计算机技术与发展,2019,29(08):47-52.[doi:10. 3969 / j. issn. 1673-629X. 2019. 08. 009]
 GUO Lei-lei,YU Lu,DUAN Guo-lun,et al.A Multi-feature Fusion Method Based on AP Clustering[J].,2019,29(08):47-52.[doi:10. 3969 / j. issn. 1673-629X. 2019. 08. 009]
点击复制

基于 AP 聚类的多特征融合方法()
分享到:

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:
29
期数:
2019年08期
页码:
47-52
栏目:
智能、算法、系统工程
出版日期:
2019-08-10

文章信息/Info

Title:
A Multi-feature Fusion Method Based on AP Clustering
文章编号:
1673-629X(2019)08-0047-06
作者:
郭蕾蕾1 ;?俞 璐1 ;?段国仑2 ;?陶性留1
1. 陆军工程大学 通信工程学院,江苏 南京 210007; 2. 陆军工程大学 指挥控制工程学院,江苏 南京 210007
Author(s):
GUO Lei-lei 1 ;?YU Lu 1 ;?DUAN Guo-lun 2 ;?TAO Xing-liu 1
1. Institute of Communications Engineering,Army Engineering University of PLA,Nanjing 210007,China; 2. Institute of Communications Control Engineering,Army Engineering University of PLA,Nanjing 210007,China
关键词:
AP 聚类;?多特征融合;?视图(特征)不平衡;?成对约束;?相似度矩阵
Keywords:
AP clustering;?multi-feature fusion;?view(feature) imbalance;?pairwise constraints;?similarity matrix
分类号:
TP391
DOI:
10. 3969 / j. issn. 1673-629X. 2019. 08. 009
摘要:
经典的聚类方法通常只适用于单一特征数据,对于多特征数据,特征融合显得尤为重要。 传统的多特征融合方式易造成维数灾难、尺度较小的特征被忽视等问题。 对于“视图(特征)不平衡”数据,上述问题显得尤为突出。为此,提出了一种基于成对约束的多特征融合 AP 聚类算法。 该算法用“差特征”数据聚类得到约束信息,利用“好特征”数据得到基础相似度矩阵,再利用成对约束来调整基础相似度矩阵,在新得到的相似度矩阵上进行 AP 聚类。 该特征融合方法中,“好特征”占据主导,“差特征”只是以约束的形式发挥作用,克服了现有特征融合方法中效果差距很大的特征平起平坐的缺点。实验结果表明,相较于单视图聚类、多视图数据直接拼接后再聚类、多视图谱聚类等方法,多特征融合 AP 聚类算法取得了较好的性能,有效地解决了“视图(特征)不平衡”问题。
Abstract:
The classical clustering methods are usually only applicable to the single feature clustering,and for multi-feature data,feature fusion is particularly important. The traditional multi-feature fusion is easy to cause dimension disaster and the small scale feature neglected. This is particularly true for unbalanced data. Therefore,a multi-feature fusion AP clustering algorithm based on pairwise constraints is proposed. The algorithm first clusters the “bad feature” data to get the constraint information and uses the “good feature” data to get the basic similarity matrix,and then use the pairwise constraints to adjust the basic similarity matrix. Finally the AP clustering on the new similarity matrix is carried out. In the process of feature fusion,“good feature” dominates the similarity,while “bad feature” only plays a role in the form of constraints,overcoming the disadvantages of the features with a large effect difference equality in the existing feature fusion methods. Experiment shows that compared with single-view lustering,multi-feature data re-clustering after splicing, multi-view spectral clustering,the proposed multi-feature fusion AP clustering algorithm can achieve better performance and effectively solve the problem of “view (feature) imbalance”.

相似文献/References:

[1]聂建豪,李士进. 基于图像识别的秸秆焚烧事件检测[J].计算机技术与发展,2017,27(05):69.
 NIE Jian-hao,LI Shi-jin. Detection of Straw Burning Event Based on Image Recognition[J].,2017,27(08):69.
[2]龚安,费凡.基于多特征融合的评论文本情感分析[J].计算机技术与发展,2018,28(08):91.[doi:10.3969/ j. issn.1673-629X.2018.08.019]
 GONG An,FEI Fan.Comment Text Sentiment Analysis Based on Multi-feature Fusion[J].,2018,28(08):91.[doi:10.3969/ j. issn.1673-629X.2018.08.019]

更新日期/Last Update: 2019-08-10