Building a Dashboard from Unstructured Sentiment Analysis of Hotel Reviews

This covers a method for collecting reviews from hotel booking platforms like Hotels.com, performing unstructured sentiment analysis and topic extraction in one pass, and displaying the results as a dashboard. Using Jurassic-1 from AI21 Studio, you can convert review text into structured JSON with almost no NLP knowledge.

The core flow is as follows.

• Collect reviews for each hotel page by page using RapidAPI's Hotels.com reviews API. Calling the reviews API requires a RapidAPI key, which is separate from the AI21 Studio key.
• The example uses the Empire Hotel, gathering only the summary of each review from the response and combining about 50 reviews per page into the full set.
• langdetect is used to filter out reviews that aren't in English or are too short, improving input quality.

The model uses a few-shot prompt that was tested and refined in the playground. The examples should cover a variety of items that frequently appear in real reviews, such as location, room, cleanliness, facilities, service, view, and AC, and the article recommends at least 8 examples. Since accuracy matters, temperature=0 is used, and the output is received in a form like {"Location": "Positive"}.

In the aggregation step, the results are parsed using ast.literal_eval and pandas, responses with broken formatting are excluded, and the Positive/Negative counts for the top 7 topics are tallied and visualized. The visualization is built based on a matplotlib horizontal bar chart example, and the dashboard shows at a glance strengths such as location, service, and cleanliness, as well as improvement points such as WiFi, AC, and rooms/facilities.

Finally, performing a simple validation by comparing against Booking.com's category ratings lets you check whether the dashboard results align with actual user evaluations. This results in a practical analysis tool that lets hotel operators quickly identify strengths and weaknesses without having to read through long reviews themselves.