I have a problem about hate-speech classification using support-vector machine algorithm. The task is to identify the sentence that contains 'positive' or 'negative' sentiment. Which is the best Kernel Trick? ('rbf' or 'polynomial')
This looks like it is a duplicate, thought not a complete one, of the following stack
In there is a link to a guide that states, starting on page 12, three scenarios to consider when choosing between a kernel or a linear method:
- Number of instances < number of features
- Both numbers of instances and features are large
- Number of instances > number of features
To summarize, linear is suggested for when the number of features is large or at least larger than the number of instances. Any other instance, a kernel would be suggested.
I also found a link to good primer I deciding between linear and kernel here.
You can condense the advice to the fact that when using SVM decide on the simplest approach first (linear) and if that does not work use RBF as polynomial does not tend to offer any performance improvements above using RBF.