The assignment a the end of the section “6 - Testing with your Own Input” has the following comment that:
Notice that the model works well even for complex sentences.
I shall admit, this model get the feedback “This movie was almost good.” right, but it fails on:
- This movie was not good => Predicted as positive
- This movie was not bad => Predicted as negative
- I laughed, it was funny => Predicted as negative
So, indeed, it caught some of complex sentences right and works better that Naive Bayes, but it looks that it still far from working well on complex sentences…