Extracting Shared Subspace for Multi-label Classification
Shuiwang Ji, Lei Tang, Shipeng Yu, and Jieping Ye
Proceedings of the Fourteenth ACM SIGKDD International Conference on Knowledge Discovery and Data Mining
The codes in this package are written in MATLAB. Please read the README file in the LS_ML.zip package first. The package requires the LIBSVM MATLAB interface and the MOSEK optimization package. For LIBSVM, you need to download the MATLAB interface, compile it if necessary, and put the compiled files into your MATLAB path. For MOSEK, you can download the installation file and get a trial license for free if no commercial purpose is involved. Once you have these two pieces of (FREE) software installed, you can easily replicate the experiments described in the paper.
The original Yahoo! data sets are available at: http://www.kecl.ntt.co.jp/as/members/ueda/yahoo.tar.gz
Note that the original data are in the term-frequency (TF) format, and we have converted them into the term-frequency inverse-document-frequency (TF-IDF) format, and all documents are normalized to unit length.
For your convenience, we have included the preprocessed Science data set with this package.
If you have any comments or questions, please feel free to contact Lei Tang and Shuiwang Ji