A quotient space reduction for large-scale support vector machine datasets

Againsting the low efficiency of training on large-scale support vector machine, use quotient space method to deduce a new sample reduction method, quotient space reduction(QSR). Unlike the traditional reduction methods, QSR brings the concept of “Granularity” into reduction method. It has two stages: CGR and FGR. In CGR, use KDC reduction to build the quotient space of original problem set with coarse grain-size. In FGR, build another quotient space with more fine grain-size. By changing the reduced intensity according to the number of redundant points remained, QSR gives a higher compression ratio and accuracy. Apply the new method to the large scale SVM social spam detection model. The detection model speed up obviously.