Happy Hotel
You've just joined the data team at a large hotelling chain. Your specific team is embedded in the CX (customer experience) division. Each day you receive hundreds of reviews of your 10 hotels from your customers. Each review consists of a free-form text review and a report of "happy" or "not happy". A product manager on your team wants to understand each hotel's performance at a more granular level; they want to build a product to identify topics within reviews. All of your reviews are unlabeled, and it's infeasible to label them by hand.
Design and execute a method to identify topics within the reviews. For each topic, find a robust means of assigning a score to each hotel in that topic. For each of the ten hotels, what specific-to-that-hotel recommendations would you make to their general managers? Engineering Challenge: How would you design this system to update over time? How would you persist topics from one timestep/update to another? How would you design your scores so they can meaningfully understand when they’re doing better?