Hi
So while using Logistic regression to classify our tweet as positive or negative, can we also classify it as neutral if value is near 0.5?
We could but that would be a “hack”. We would have to “look” at the output distribution and decide “how much near 0.5” the tweets should be considered neutral. Also, there would be no direct way to train this label.
The common way is to have multivariate output (multiclass classification) and softmax applied on the outputs. This way the three categories’ probabilities would sum to one and you could train the network for this type of added label.
Cheers
1 Like
For logistic regression or classification, classes are always True or False, or are specific enumerated labels.
When there are multiple classes, the one with the highest value is declared to be the best prediction.