A Simple Hotel Recommender - WIP

A fun evening (or three) project to toy around with LLM APIs. When searching for a hotel, the process of sifting through numerous reviews to decipher the overall sentiment is tiresome and inefficient. A natural language interface can greatly improve this experience, acting as a human like travel agent. This is a POC for an app that receives a location as input and generates a paragraph that summarizes and synthesizes the key points from various reviews. Try it yourself!

Implementation Details

• Source code is available in this HF space.
• The app uses data from the CMU Hotel Review Dataset, consisting of almost 1M reviews crawled from TripAdvisor.
• A (random) hotel is recommended if it satisfies the following:
  • Its average score is in the top 80% for that city
  • It has more than 100 reviews
• The app generates a paragraph for the recommended hotel based on four random reviews. Only reviews that were found helpful by at least 10 users are considered.
• Three of the four reviews are positive (score = 5), and one is negative (score <= 2).
• The LLM is asked to generate a compelling yet honest argument.