They ask: “How good is this restaurant, on average?”
We ask: “How much will you like it?”
There is no global score here—because global scores are meaningless. A tasting-menu temple and a minimalist bistro don’t compete on the same axis, and neither benefits from being flattened into a single number.
Instead, this platform works more like Netflix’s recommendation engine than Yelp.
On most review sites, people feel pressure to “be nice.”
Five stars here, four stars there. Everything blurs together.
That helps no one.
Rate the Pittsburgh fine dining spots you've visited. Be brutally honest — this isn't about being nice or fair to restaurants. It's about capturing your genuine taste preferences so we can find your culinary soulmates.
Behind the scenes, we build a correlation matrix that maps how similarly different reviewers rate restaurants. If you and another reviewer both loved Altius but were underwhelmed by the menu at Morcilla, you're probably going to agree on other places too.
The clever part: we use transitive relationships. Even if you and another reviewer haven't rated the same restaurants, we can infer your likely agreement through chains of similar tasters. Think of it as "friends of friends" for your palate — the system discovers hidden connections between taste preferences across the entire network.
No restaurants yet
How closely other users' ratings correlate with yours
Configure the recommendation engine parameters. Changes take effect immediately.
Controls how much to discount user influence based on shared reviews.
The influence formula is: influence = correlation × (n / (λ + n))
where n = reviews in common. Higher lambda means more shared reviews are needed
before a user's correlation fully impacts recommendations.
Low values (1-3): Trust correlations quickly, even with few shared reviews.
High values (10+): Require many shared reviews before trusting correlations.
Number of latent factors used in matrix factorization for computing transitive correlations.
This determines how many underlying "taste dimensions" the system uses to find similar users.
Low values (1-2): Simpler model, captures only major taste patterns.
Higher values (5+): More nuanced model, can capture subtle taste differences.
Note: Changing K triggers a full engine recompute (O(R³) where R = users).
Maximum number of users that can register. New registrations will be blocked when this limit is reached.
Maximum number of restaurants that can be added. Adding new restaurants will be blocked when this limit is reached.