Multi class text classification when having only one sample for classes

Question

I have a dataset of texts, each text was identified with an ID number. I would like to do a prediction by finding the best match ID number for upcoming new texts. To use multi t