How the score is built

Booking.com asks guests two separate questions: what they liked, and what they disliked. That split is the whole trick. A complaint about sound lands in the disliked field already labelled negative, so counting them needs no sentiment model and no guesswork.

The inputs

For each hotel we read 283 English-language verified-stay reviews and match both fields against a deliberately narrow word list — noise, noisy, loud, thin walls, earplugs, couldn't sleep, kept us awake, sirens and a handful more. Narrow on purpose: missing a complaint is better than inventing one.

The formula

Start at 100. Subtract 2.5 points for every percent of guests who raise noise. Subtract 1 point per percent who complain about the bed. Add 1 point per percent who praise how restful the room was. Clamp to 0–100.

Noise is weighted heaviest because it is the only dimension that actually separates properties. Across every city we have scored, noise complaint rates span 0% to 30.7%. Bed complaint rates span a narrow band that tells you very little.

What this does not do

It does not measure decibels. Nobody stayed in these rooms with a sound meter. It measures how often paying guests were bothered enough to write it down, which is a different thing and arguably the more useful one.

It also cannot see room-level variation. A hotel can be restful on the courtyard side and loud on the street side, and the score flattens that into one number.

Honest limits

This is early data: 55 hotels across 4 cities, 15579 reviews. At this sample size the gap between the best and worst properties is solid. The ordering of the middle is not — a two-point difference there is four reviews, which is chance.

Reviews are quoted in short excerpts as evidence for the score, never reproduced in full.