First thing: There is no difference when an SVM is used for text classification with regard to its internal mechanisms.
You already grasped that the Linear Kernel is well suited for text classification. The Linear Kernel is computationally very cheap (as opposed to many other Kernels) and usually works well for text classification problems.
Assuming you know a little about SVMs and know how to apply them, you just need to transform your texts in a suitable representation for the SVM.
Take at look at this paper:
- look at the feature representation (bag of words, not mentioned in the paper explicitly if i remember correctly)
- the weighting methods may interest you (for example tf-idf)
- skip the feature selection at the beginning (although its in the paper)
I hope this gives you a starting point for using text classifiation.