Build and iteratively improve sentiment classifier

You need to build a sentiment classification model (e.g., positive/neutral/negative) for user-generated text. You already shipped an initial version, and the interviewer asks a project deep-dive. Explain: - How you formulated the problem (labels, classes, unit of prediction, multilingual/emoji handling). - Why you chose your modeling approach (baseline vs deep model) and what alternatives you considered. - Your data pipeline and labeling strategy (human labels, weak supervision, distant labels, class imbalance). - How you evaluated the model (metrics, train/validation split, leakage risks) and what error analysis you did. - How you iteratively refined the system based on findings (data cleaning, feature/model changes, thresholding, calibration). - What you learned during iteration and what you would do next.