I came up with a modified version of TF-IDF function for text retrieval task. I want to do retrieval experiments using Vector Space Model and compare my function to some of those proposed in the literature in terms of known metrics (such as MAP and interpolated precision at recall...etc).
In literature, there are many improved TF-IDF functions that where proposed for text retrieval task and text classification task. Is it feasible to compare my function to those that were originally used for classification?
I know that the idea of representing documents as vectors of terms weights is the same in both cases. But, can I simply take a function that was mainly developed for improving classification, and apply it in a retrieval process to compare its results to my function's results? Of course, I don't mean those functions that use "class"-related or "category"-related features.