I have a dataset that contains around 10000 small paragraphs and the paragraphs belong to classes. There are around 80 - 100 classes. The paragraphs can be organized in hierarchies. I want to build a classifier model that will predict the class of an unseen paragraph.
Currently what I have done is, I have implemented a two step classification using FastText. First I classify the unseen text to a top level class and then using another classifier I classify it to a subclass of the identified top level class. This helped me to increase the accuracy.
Is there a better way to do this? Is there any good hierarchical classifier like https://github.com/globality-corp/sklearn-hierarchical-classification out there for text classification? Or can this be improved using FastText itself in some way?