Plan a Topic Discovery Project on 50,000 Customer Messages

A product team has 50,000 free-text feedback messages (median 25 words) and no taxonomy. They ask: "What are people talking about, and can we track it every month?"

1. Propose a plan from raw text to a monthly dashboard, naming the models you would try.
2. Compare LDA and a BERTopic-style pipeline for this data, and say how you would choose the number of topics or clusters.
3. Describe how you would evaluate topic quality when there is no ground truth.
4. Explain how and when you would convert the discovered topics into a supervised classifier.
5. List three failure modes of topic models in a monthly tracking setting.