How TheKnow works
The technology behind personalized restaurant recommendations, and how it differs from a crowd average.
The problem with crowd reviews
Review sites average ratings from thousands of strangers. A restaurant's score is a single number meant to represent everyone.
But taste is personal. The regular who comes for the late-night burger and the couple in for an anniversary rated the same place, and their opinions got blended together. The resulting 3.8 tells you almost nothing about whether you will enjoy it.
Averages can also be pushed around: fake reviews, rating campaigns and a burst of early fans all move the number. That new place with 4.9? It may just not have enough ratings yet.
A 4-star average might be a 2 for you, or a 5. The crowd can't tell you which.
Our approach: taste matching
Instead of showing you what everyone thinks, TheKnow finds the people who rate like you, and shows you what they love.
Think of a music app that plays you what similar listeners love. TheKnow does that for restaurants.
When you rate, we compare your ratings with everyone who rated the same places, and look at two things. Pattern: when you rate a place high, do they? When you rate one low, do they? Closeness: are your scores within half a point? People who clear the bar are your taste matches, and the closest of them are your taste twins. Together they're your people.
The algorithm explained
Three key concepts power TheKnow's taste matching engine.
Match = pattern plus closeness
Pattern is Pearson correlation: do your ratings move together? Closeness is the share of places where your scores are within half a point. Your match percentage is the average of the two. It says how alike you rate: you rank places the same way and score them close.
The bar
Nobody becomes a taste match on fewer than 5 places in common. Past that, the pattern has to hold up: we count it only when we're 80% sure it isn't chance. People who score alike can also qualify by agreeing within half a point on 60% of at least 8 shared places. Anyone who doesn't clear the bar isn't shown as a taste match.
A taste twin rates like you: the same places at the top and bottom, and close scores on most of them. It's the strictest kind of taste match: at least 8 places in common, a strong pattern, and close scores on 60% of them.
Evidence
A taste match on 20 places means more than one on 6. Each taste match says how much it rests on: early (5 to 7 places), growing (8 to 14) or solid (15 or more). More places in common lifts a taste match up your list and gives it more weight in your people's read.
Your people vs a crowd average
Same restaurant, two ways to read it.
| Feature | A crowd average | Your people |
|---|---|---|
| Whose ratings | Everyone who rated it | Your taste matches and twins, people who rate like you |
| Same for everyone | Yes | No, yours alone |
| How it's built | An average | Pattern plus closeness, weighted by how alike you rate |
| Best for | A general reputation | Finding places you'll love |
Privacy and transparency
Your ratings power the matching, and your data stays yours. We never sell it. Your ratings find your taste matches and shape recommendations; public ratings also count toward a restaurant's rating count.
Can other users see my ratings?
Your ratings start public. You can make any single rating private when you rate it or later: nobody else sees it, and it isn't used to find your taste matches. To take all of your ratings off your profile, turn off "Show my ratings publicly" in Settings. Those ratings still help find your taste matches, who then see how closely you agree, but not what you rated, your scores or how many places you've rated.
We believe in transparency. The algorithm isn't a black box: this page explains how it works. If you have questions about your data, our privacy policy has the details.
Ready to try it?
Join TheKnow and start getting restaurant recommendations from people who actually share your taste.