SVMmap on LETOR 

　
Introduction

Learning parameters

Papers&Docs

Notes

　

Introduction to SVMmap
SVMmap is a structural Support Vector Machine method which optimizes for an upper bound of average precision loss in the predicted rankings.  

The details of SVMmap can be found at http://projects.yisongyue.com/svmmap/ 

The details of structural SVMs can be found at http://www.cs.cornell.edu/People/tj/svm_light/svm_struct.html

　

Learning Parameters

We used the package available at http://projects.yisongyue.com/svmmap/.  The C parameter was set using the validation set.

Dataset

-c (from Fold1 to Fold5)

OHSUMED

0.00001, 0.8, 2, 2, 10

TD2003

100000, 1000, 0.1, 0.1, 1000000

TD2004

100000, 100, 1000000, 10000, 100000

HP2003

10000, 0.05, 0.01, 0.04, 0.03

HP2004

4000, 1000, 50000, 2000, 9.5

NP2003

0.09, 10000, 30000, 40000, 100000

NP2004

0.08, 60, 100000, 4.5, 70

Papers & Docs

Y. Yue, T. Finley, F. Radlinski and T. Joachims. A Support Vector Method for Optimizing Average Precision, Proceedings of the ACM Conference on Research and Development in Information Retrieval (SIGIR), 2007

I. Tsochantaridis, T. Joachims, T. Hofmann, and Y. Altun, Large Margin Methods for Structured and Interdependent Output Variables, Journal of Machine Learning Research (JMLR), 6(Sep):1453-1484, 2005.

 

BIBTEX

@InProceedings{Yue/etal/07a,
    author = {Yisong Yue and T. Finley and F. Radlinski and T. Joachims},
    title = {A Support Vector Method for Optimizing Average Precision},
    booktitle = {ACM Conference on Research and Development in Information Retrieval (SIGIR)},
    pages = {271--278},
    year = 2007
}

@Article{Tsochantaridis/etal/05a,
    author = {I. Tsochantaridis and T. Joachims and T. Hofmann and Y. Altun},
    title = {Large Margin Methods for Structured and Interdependent Output Variables},
    journal = {Journal of Machine Learning Research (JMLR)},
    year = 2005,
    volume = 6,
    pages = {1453--1484},
    month = {September}
}

 

Notes
Experiments conducted by Yisong Yue.  If any problem, please contact letor@microsoft.com