Presentation | 2011-03-28 A Study on Position-based Adaptive Weighting for Ranking SVM Masayuki KARASUYAMA, Takuya HASEGAWA, Tsukasa MATSUNO, Ichiro TAKEUCHI, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | This paper presents a novel training algorithm for ranking support vector machine (ranking SVM). The focus is on how to incorporate position-dependent information in weighted variant of ranking SVM. It has been recognized in information retrieval literature that quality measures (such as normalized discounted cumulative gain: NDCG) for ranking should incorporate position-dependent information because users are only interested in a few top-ranked documents. An important issue in this task is that the final position of each document, i.e., ranking result, is unknown before training the ranking SVM. Our approach in this paper is adaptively changing the weights in the training process based on the position information produced by the ranking SVM currently under training. Specifically, we introduce regularization path-following approach and extend it in such a way that position-dependent weights are adaptively updated by detecting position changes in the regularization path. We present two implementations of this approach: exact one and approximated one. We show that the former can exactly keep track of the position changes, while the latter can be much faster and more scalable than the former by approximating the regularization path. We demonstrate the effectiveness of our approach by applying it to large-scale information retrieval task. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | Learning to rank / Ranking SVM / adaptive weighting / path following |
Paper # | IBISML2010-115 |
Date of Issue |
Conference Information | |
Committee | IBISML |
---|---|
Conference Date | 2011/3/21(1days) |
Place (in Japanese) | (See Japanese page) |
Place (in English) | |
Topics (in Japanese) | (See Japanese page) |
Topics (in English) | |
Chair | |
Vice Chair | |
Secretary | |
Assistant |
Paper Information | |
Registration To | Information-Based Induction Sciences and Machine Learning (IBISML) |
---|---|
Language | JPN |
Title (in Japanese) | (See Japanese page) |
Sub Title (in Japanese) | (See Japanese page) |
Title (in English) | A Study on Position-based Adaptive Weighting for Ranking SVM |
Sub Title (in English) | |
Keyword(1) | Learning to rank |
Keyword(2) | Ranking SVM |
Keyword(3) | adaptive weighting |
Keyword(4) | path following |
1st Author's Name | Masayuki KARASUYAMA |
1st Author's Affiliation | Department of Engineering, Nagoya Institute of Technology:Research Fellow of the Japan Society for the Promotion of Science() |
2nd Author's Name | Takuya HASEGAWA |
2nd Author's Affiliation | Department of Engineering, Nagoya Institute of Technology |
3rd Author's Name | Tsukasa MATSUNO |
3rd Author's Affiliation | Department of Engineering, Nagoya Institute of Technology |
4th Author's Name | Ichiro TAKEUCHI |
4th Author's Affiliation | Department of Engineering, Nagoya Institute of Technology |
Date | 2011-03-28 |
Paper # | IBISML2010-115 |
Volume (vol) | vol.110 |
Number (no) | 476 |
Page | pp.pp.- |
#Pages | 7 |
Date of Issue |