If I first classify an intent into classes using SVM classifier and then within those classes I classify that intent into subclass using another SVM classifier , will it be helpful or overall accuracy will remain the same?
1 Answer
$\begingroup$
$\endgroup$
2
My bet is, go for the entire list of classes whatever you can.
Treat this as a multi-label classification approach. You can get more than one result. If you are using deep learning, you can opt for sigmoid as activation function in the output layer instead of softmax. Can't say for sure whether your accuracy will increase for the deep learning model, depends on the quality and quantity of data.
-
$\begingroup$ The subclasses are all similar to each other for each class. Still should I go ahead with all classes and data is very limited ...faqs $\endgroup$ Jun 27, 2018 at 8:49
-
$\begingroup$ How limited? Try deep learning only when you have at least 1000 observations per class, check the toxic comment classification challenge on Kaggle. Seems to suit your needs. Multilabel classification with classes which are actually close to each other. $\endgroup$– tenshiJun 27, 2018 at 8:55