Poor result of sentiment analysis

I am in the very first course of fast ai course and trying all the sample code given , when trying out NLP sample , I noticed the following that it predicts the sentence with explicit negative text as positive review !!! but the model accuracy is 92%

can any one explain what’s going wrong here , data insufficiency or overfitting ?

I think this example may be kind of iffy.

You can consider 'too bad !' as 'ah, too bad, better luck next time, it was a good try!'

Maybe have another go at learn.predict but try 'That was terrible.'

Other then that, I have no idea, the model looks fine to me


1 Like