Classifying text using machine learning models and determining conversation drift