InthisstudyweproposeamethodforfeatureselectionanditerativeclassifiertrainingbasedonSupportVectorMachines(SVM)withlinearkernels.Weexplorehowthisandotherfeatureselectionmethodscanbeusedtomaketradeoffsbetweentheamountoftrainingdataandthesparsityofthedocumentrepresentationforthefixed