Is there a ML Algorithm to map labels (Single or max a couple of words) to description (a text telling more about a label)

Question

So, I have a metadata dataset with labels and their descriptions. A sample from the dataset looks like the following:

Label	Descriptions
Release Date	Date of formal issuance
Language	The language of the dataset

I want to train a ML model which learns the relationship between Label (Input X) and Descriptions (Target Y-categories) and then, can predict the category or the description of an unseen Label from the given list of Categories / Descriptions. Here we assume, the new label would be similar, in spelling or in meaning, to one of the labels used in the training model.

Unfortunately, most of the algorithms try to map the description (which is usually a text document, review etc) to one of the categories (positive or negative etc)

Would be great to get some help here, as to which algorithm would help me solve this problem.

Thanks in advance!!

Answer 1

I don't think it is possible: it can't be framed as a classification task nor a translation/transformation one, in fact at a high level the description is a better explaination of the label, tailored with external (domain?) knowledge that cannot be expressed in any model I know of.

Besides that, I don't think you have the necessary data amount and variability to express a sufficent generalization over the output.