I would like to know if there is a performance improvement in pre-processing the user input somehow. For example, removing stop-words, stemming the phrases, lemmatizing the sentences, etc...
We have been testing the service and sometimes it makes mistakes in different tenses in spanish (past tense, perfect tense, etc...) that we test in the test set and are not trained and the actual intent. The intention tested is the same that the one that is actually trained
I would like te get a card
I am trying to get a visa card
I am interested in mastercard
How can I get a credit card?
etc...
But with the training made above, it makes mistakes at the sentence "It would have been interesting to get a card".
Do you guys have any advice in the training or pre-processing task to perform better at not seen examples?