I am working on a project where I have to do multi-label text classification. I want to understand that whether my approach is correct or I am missing something. I am using R to do it.
- Clean the text
- Create a corpus. While creating the corpus I am removing the sparse term with sparse value set to .86
- Create a DTM from this corpus and attach the label to it.
- Divide the DTM into training and test set
- Build the model using the train set
- Test the model using the test set.
- Do statistical analysis to prove the model
So my questions are
- Do we have to remove sparse term while creating the corpus and if so does the sparse value is too high and is there any range of sparse value to be set in text classification or it depends on the result we get.
- Is it a thumb rule that the number of rows should be more than the square of the number of features. And also does number of features depend on the number of unique classes you have
- Can we do feature selection once we do sparse term removal.
- Which one gives good result dimension reduction or feature selection.