LibSVM vs. LibShortText

Question

What's the difference between using a linear classifier with LibSVM vs. using LibShortText or LibLinear, if there is a difference at all? The same goes for using SVMlight; are they implemented differently?

Thanks.

Answer 1

LibShortText is built upon LibLinear and specified in short-text classification.

LibLinear is considered faster than linear LibSVM, and often used for large scale data set. Prof. Lin's website shows When to use LIBLINEAR but not LIBSVM. The result from LibLinear and LibSVM are considered close to each other when you set '-b 1' in LibLinear (note that the bias term is not used by default). The difference between the results even with the added bias is supposed to be due to the regularization term (w'w+b^2).

SVMlight is also a good option in large data set classification, and it is especially efficiency in implementation of the leave-one-out cross-validation.