I am familiar with
SMOTE (Synthetic Minority Oversampling), and the Python Library
imbalanced-learn, which can be used to handle minority samples in a data, but with numeric format only. I am currently facing a dataset, in CSV format, which has amongst many other columns, a final column, say named
Target, which contains brief sentences in every row (1 sentence per row i.e. the target is a string, unlike numeric data/predicted class). But, the problem is, that majority of these sentences are all exactly same, and only a few of them are different. This is the first problem, though, all rows have the target as a sentence. I wonder if there is any technique to oversample/create more examples for my data which handles the minority sentences/targets.
I would like to use my other columns (all numeric) to predict the target, which, in this case happens to be a
string (sentence). Basically, we can think of this as something like a classification problem, where the targets/responses are
Strings instead of
classes. I am very confused regarding how to accomplish this, and even if this may not give good results necessarily, I would like to know of an approach to do this. If anyone can suggest a sample (baby example), or point to a real world coding example in Python which handles a task of this type, it would be much appreciated. Cheers!