Improving Multiclass Text Classification with the Support Vector Machine

Papers

• Jason Rennie and Ryan Rifkin. Improving Multiclass Text Classification with the Support Vector Machine. Massachusetts Institute of Technology. AI Memo AIM-2001-026. 2001. Also available as PDF.

Code/Matrices

• We used Rif's SvmFu for training SVMs.
• We used Rainbow for pre-processing text documents.
• A package, ecoc-svm-code.tar.gz (54k), includes additional Java and Perl code we wrote, in addition to the matrices we used.
• Rayid Ghani hosts a package of various-size BCH codes [local].
• A list of SMART word vector weightings.

Data

• 20 Newsgroups (we used the 19997 version)
• Industry Sector
• A (large) package, ecoc-svm-data.tar.gz (99M), contains the pre-processed data we used.

Experiments

The ecoc-svm-code.tar.gz package includes instructions on how to run the experiments. Make sure that the code and data packages are unpackaged into the same directory.