Scenario

You are working on an NLP system and need to turn raw text into features that a model can learn from. The task could be something like document classification, routing, or tagging, and the main challenge is choosing representations that capture useful signal without adding unnecessary complexity.

Question

How do you approach the problem of feature extraction in natural language processing?

Anchor your answer in a realistic text classification problem, such as routing Adobe Experience Manager support tickets into operational queues. Discuss how your feature choices change from sparse lexical baselines to dense semantic representations.