What classification models can I use for a task with 10k classes?

I am doing this task predicting roughly 10 thousand classes, it is an author identification task, 10k twitter users !!! I’m wondering what deep learning models are suitable for such task? I know CNN is good for classification but I am not sure if CNN can really perform well in 10k classes ? I’m also thinking about machine learning models like KNN, SVM, Naive Bayes, I will try all these models, but this is the first time that I deal with such much classes, anyone has other suggestions ? Thank you


If it’s text based, ULMFiT is one of your best options to go with. See the IMDB notebooks to see how it is done.

1 Like