Am I missing something, or does the word ‘adore’ actually appear in the training set (contrary to what’s said just above in the section headed The Model Matches Emojis to Relevant Words)?
Running the below shows that ‘I adore my dogs’ is the 42nd sentence and has label 0 (i.e. the correct positive label). Unless I’m mistaken this is likely just an oversight - but could easily be fixed by changing the word to ‘cherish’ which is the same meaning, gives a heart emoji and isn’t in the training set.
for i, stc in np.ndenumerate(X_train):
if(stc.find('adore') != -1):
print('Sentence', i[0],'is ',stc, 'with label =', Y_train[i])