Viator Data Scraping

Review volume is part of discovery, not just quality

On a pure marketplace, reviews are a quality signal a buyer reads after finding a listing. Here the platform sits inside a review ecosystem, and review volume participates in how a listing is found at all.

That has consequences for what the data shows:

• A long-established activity with many reviews has a discovery advantage independent of price.
• A new listing competes differently, regardless of quality.
• So position and review count correlate, and reading position as a demand signal conflates the two.
• Review counts accumulate, so they are a stock rather than a flow.

We deliver review_count and rating alongside position and the query that produced it, so the correlation is visible. We do not model a ranking algorithm, and we do not present position as a demand measure.

Review counts are a stock

A count of 4,000 reflects years of accumulation, not current volume. review_count_delta against the prior batch is the flow measure, and it is the more informative one for anything time-sensitive.

The timeslot model, applied

Everything our GetYourGuide page sets out applies here and we apply it rather than restating it at length:

• The record is a departure timeslot on a date, not a product. Slots price and sell out independently.
• Price tiers sit beneath the slot — adult, child, group — each its own record.
• Language is a dimension, not an attribute.
• The operator sets the price, and inclusions vary between operators.
• Cancellation windows are structured fields, since a flexible booking and a restricted one are different products.
• Inclusions stay as published text with inclusions_parsed: false.

Cross-platform operators

The same operator frequently lists the same activity on more than one platform, at different prices because commission differs. also_on_other_platform is recorded where both are in scope and the gap computed from paired records.

We do not match activities across different operators as equivalent. Two walking tours of one site can differ in group size, guide language, entry inclusion and route.

What we do not collect

• Reviewer identities or review text. Counts and ratings only — and this matters more here, where reviews are central to the product.
• A ranking model. We record position with its query; we do not infer how it was produced.
• Position as a demand measure. It correlates with review stock as much as with demand.
• Bookings, revenue or places sold. Not published.
• Capacity from a scarcity message. That is merchandising, not an inventory count.

On review text specifically

Activity reviews frequently name guides, describe other participants, and contain personal detail. We do not collect review text, and where a client wants sentiment we say that it would require collecting text we decline to take.

Counts and ratings support most of what sentiment analysis is actually asked for — whether an activity is well-received and whether that is changing.