Abstract | ||
---|---|---|
We investigate the problem of data-driven ordinal regression—the problem of learning to rank order new data items based on
information inherent in existing data items. Ordinal regression shares common features with multi-category classification
and metric regression. However, it requires new, tailor-made methodologies to reduce prediction error. The approach has application
in various domains, including information retrieval, collaborative filtering and social sciences. We propose a new distribution-independent
methodology for ordinal regression based on pairwise preferences employing information about strength of dependency between
two data instances, which we refer to as continuous preferences. Our hypothesis is that additional information about strength of preference as well as its direction can improve algorithmic
performance. We also provide a novel technique for deriving ordinal regression labels from pairwise information. Experimental
results on real-world ordinal and metric regression data sets confirm usefulness of the methodology compared with other state-of-the-art
approaches. |
Year | DOI | Venue |
---|---|---|
2012 | 10.1007/s13042-011-0036-x | Int. J. Machine Learning & Cybernetics |
Keywords | Field | DocType |
Ordinal regression, Pairwise preference, Preference training, Continuous preference | Learning to rank,Pairwise comparison,Data mining,Collaborative filtering,Ordinal number,Ordinal data,Ordinal regression,Preference learning,Ordinal optimization,Mathematics | Journal |
Volume | Issue | ISSN |
3 | 1 | 1868-808X |
Citations | PageRank | References |
3 | 0.41 | 13 |
Authors | ||
3 |
Name | Order | Citations | PageRank |
---|---|---|---|
Maria Dobrska | 1 | 3 | 1.43 |
hui wang | 2 | 76 | 17.01 |
William Blackburn | 3 | 9 | 4.92 |