How should we handle the mispelled words in a NLP problem, suppose I would like to do sentiment analysis on product reviews but many words are mispelled in the reviews. What are possible ways of handling this scenarios
Possible ways to handle them should be:
a) Two fix them properly so the model can use them for learning
b) to make then unknown tokens but can be used for learning
If the misspelling is same for similar words they could be used for training(learning) but the problem is when testing the model in real life you will have negative impact on performance.