Question: Describe how a multi-stage sentiment analysis system differs from an end-to-end learning system, using a restaurant review as the example.

Sample answer: A multi-stage system breaks the task into separate pieces. For instance, it might first tokenize the review, then extract hand-designed features, and then send those features to a classifier that predicts whether the review is positive or negative. An end-to-end system uses one learning model to go from the raw review text directly to the sentiment label. For a review such as "The soup was cold, but the dessert was excellent," the end-to-end model learns from the original text itself instead of relying on several separate processing modules.

Key points:

• Multi-stage systems use separate processing steps before classification.
• End-to-end systems learn the full mapping with a single model.
• End-to-end systems work directly from raw text.
• The final output is the sentiment label.

Rubric: The student should explain that pipeline systems use intermediate components, while end-to-end systems learn directly from raw input and produce the sentiment label without separate feature-engineering stages.