Scenario

You are working on an NLP system and need to turn raw text into model-ready features. The choice of representation affects both model quality and operational complexity, especially when moving from simple baselines to modern transformer-based systems.

Question

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

Anchor your answer in a realistic NLP problem such as classifying Adobe support messages into intents like export issue, billing issue, account access, or how-to guidance. Discuss how feature extraction changes across sparse lexical models and dense neural representations.