[1]LIU Yanglei,LIANG Jiye,GAO Jiawei,et al.Semi-supervised multi-label learning algorithm based on Tri-training[J].CAAI Transactions on Intelligent Systems,2013,8(5):439-445.[doi:10.3969/j.issn.1673-4785.201305033]
Copy
CAAI Transactions on Intelligent Systems[ISSN 1673-4785/CN 23-1538/TP] Volume:
8
Number of periods:
2013 5
Page number:
439-445
Column:
学术论文—机器学习
Public date:
2013-10-25
- Title:
-
Semi-supervised multi-label learning algorithm based on Tri-training
- Author(s):
-
LIU Yanglei1; 2; LIANG Jiye1; 2; GAO Jiawei1; 2; YANG Jing1; 2
-
1. School of Computer and Information Technology, Shanxi University, Taiyuan 030006, China; 2. Key Laboratory of Computational Intelligence and Chinese Information Processing of Ministry of Education, Shanxi University, Taiyuan 030006, China
-
- Keywords:
-
multi-label learning; semi-supervised learning; Tri-training
- CLC:
-
TP181
- DOI:
-
10.3969/j.issn.1673-4785.201305033
- Abstract:
-
Traditional multi-label learning is in the sense of supervision, in which the complete category labels are required. However, when the size of data is large and there are several categories of labels, it is quite difficult to obtain the training sample sets with complete labels. Therefore, a semi-supervised multi-label learning algorithm based on Tri-training (SMLT) is proposed. In the learning stage, SMLT initially introduces a virtual label, then for each pair of virtual labels, the Tri-training algorithm is utilized to train the corresponding classifiers for each pair of labels. In the forecast stage, a new sample is given, which will be substituted into the obtained classifier described above. According to the votes of each label, the multi-label learning problem is transformed into a label ranking problem, subsequently; the votes of the virtual label are taken as the threshold for distinguishing the label ranking results. The contrast experiments on four commonly used UCI multi-label datasets show the SMLT algorithm behaves better than other comparative algorithms in four evaluation indices and the effectiveness of the proposed algorithm is verified.