- A hotel named with its city in the local spelling is found when the stay is sold out. "Hotel Sacher Wien" was
declined ("did not match a hotel by name") for a night the hotel had no offers: Google calls the hotel "Hotel
Sacher Vienna" and writes the region in English, while its address says "1010 Wien". The address is now read as
part of the place, so the city the buyer wrote, in either spelling, is not held against the hotel. A name whose
other words do not match is still declined. What a row holds and what is charged are unchanged.
What guests say, from the page that is already read for the rate parity. No new request, no new event,
no change of price.
otherSites: the hotel's ratings on other travel sites as Google lists them (Tripadvisor, Trip.com,
Priceline and others), each with its scale, number of reviews and page.
otherSitesAverage, otherSitesReviews and ratingGap: the mean of those ratings on a five-point
scale weighted by their reviews, how many reviews that is, and Google's rating minus that mean.
praisedTopics and criticisedTopics: up to three topics Google flags as praised and as criticised,
each with at least ten mentions behind it.
- Two new
signals: weak_ratings_elsewhere (the mean elsewhere is 3.5 or below) and
rating_gap_to_google (Google's rating is 0.5 or more above it), both resting on at least 30 reviews
on the other sites.
- A Reputation view in the dataset overview. The fields are empty when the hotel's Google page was not
read: a search card carries only the first six topics and no other site.
First build.
- Hotels by search text (
searchQueries, with an optional wider search over kinds of lodging) and
by name or Google Hotels link (hotels); one row per hotel.
- From Google Hotels: stars, rating, reviews, address, phone, coordinates, and the price on every
booking site with rate parity (
parityStatus, officialUndercutBy).
- From the hotel's own site (home page and up to two contact pages, robots.txt first): e-mail
addresses sorted into role, domain and free mail, phones, social profiles, the booking engine
(about sixty recognised), the site builder and facts about the site.
- A chain's site, a booking site's or marketplace's page, a social page and a booking engine's own
page are not read; a site whose robots.txt closes its home page is not read.
signals: seven fixed rules (undercut by a booking site, no own offer on Google, only links to
booking sites, no booking engine detected, outdated site, no mobile viewport, http only).
- Filters on the search card (stars, rating, reviews, chains, vacation rentals) before anything else
is read; filters after the site is read (only hotels with an address, only hotels undercut).
- Two events:
hotel, and contact-found on top when the site gave a contact.