TY - GEN
T1 - Instance-ranking
T2 - PAKDD 2012 International Workshops: 3rd Data Mining for Healthcare Management, DMHM 2012, Multi-View Data, High-Dimensionality, External Knowledge: Striving for a Unified Approach to Clustering, 3Clust 2012, GeoDoc 2012 and 2nd DSDM 2012
AU - Xia, Xin
AU - Yang, Xiaohu
AU - Li, Shanping
AU - Wu, Chao
PY - 2013/3/5
Y1 - 2013/3/5
N2 - Single-label classification refers to the task to predict an instance to be one unique label in a set of labels. Different from single-label classification, for multi-label classification, one instance is associated with one or more labels in a set of labels simultaneously. Various works have focused on the algorithms for those two types of classification. Since the ranking problem is always coexisting with the classification problem, and traditional researches mainly assume the uniform distribution for the instances, in this paper, we propose a new perspective for the ranking problem. With the assumption that the distribution for the instance is not uniform, different instances have different influences for the distribution, the Instance-Ranking algorithm is presented. With the Instance- Ranking algorithm, the famous K-nearest-neighbors (KNN) algorithm is modified to confirm the validity of our algorithm. Lastly, the Instance-Ranking algorithm is combined with the ML.KNN algorithm for multi-label classification. Experiment with different datasets show that our Instance-Ranking algorithm achieves better performance than the original state-of-art algorithm such as KNN and ML.KNN.
AB - Single-label classification refers to the task to predict an instance to be one unique label in a set of labels. Different from single-label classification, for multi-label classification, one instance is associated with one or more labels in a set of labels simultaneously. Various works have focused on the algorithms for those two types of classification. Since the ranking problem is always coexisting with the classification problem, and traditional researches mainly assume the uniform distribution for the instances, in this paper, we propose a new perspective for the ranking problem. With the assumption that the distribution for the instance is not uniform, different instances have different influences for the distribution, the Instance-Ranking algorithm is presented. With the Instance- Ranking algorithm, the famous K-nearest-neighbors (KNN) algorithm is modified to confirm the validity of our algorithm. Lastly, the Instance-Ranking algorithm is combined with the ML.KNN algorithm for multi-label classification. Experiment with different datasets show that our Instance-Ranking algorithm achieves better performance than the original state-of-art algorithm such as KNN and ML.KNN.
KW - Instance Ranking
KW - KNN
KW - ML.KNN
KW - Multi-label Classification
UR - http://www.scopus.com/inward/record.url?scp=84874404139&partnerID=8YFLogxK
U2 - 10.1007/978-3-642-36778-6_10
DO - 10.1007/978-3-642-36778-6_10
M3 - Conference Paper
AN - SCOPUS:84874404139
SN - 9783642367779
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 112
EP - 123
BT - Emerging Trends in Knowledge Discovery and Data Mining - PAKDD 2012 International Workshops
PB - Springer
Y2 - 29 May 2012 through 1 June 2012
ER -