feature selection for text classification

negar_mahini

Dear friends
I work on text classification with reuters-21578 dataset in rapid miner. When preprocessing will complete and word vector creates and terms are weighted with tfidf, i can not use the feature selection teqniques (such as IG, MI, GINI, ...) and then validation (contains classifier and performance element). Between process documents(preprocessing) and validation, how to use feature selection?
thanks

MariusHelf

Hi,

what's the problem with using feature selection techniques? You can use them on the output of Process Documents the same way as you do on other data, i.e. a combination of Weight by XXX and Select by Weight.

As an alternative approach, you should try to train (and optimize) an SVM with a linear kernel on the complete data set with all attributes.

Best regards,
Marius