Criticker's Algorithm Explained
Criticker is home to some of the most complicated and mind-bending mathematics on the internet. We is very super smart, and have employed all of our formidable brainpower into the development of a predictive algorithm so brilliant, we suspect it might have created us, instead of the other way around.
Join us on this deep dive into Criticker's algorithm. It's intimidating, but fear not! If you can handle advanced analytic concepts like "averages" and "addition", you just might be able to follow along!
1. Rate films
1. Rate films
In order to craft its personalized recommendations, Criticker needs to know a little about your tastes. So the first step is for you to start rating films. Using a scale that ranges from 0 to 100, you need to rate at least 10 titles before Criticker can start generating recommendations for you.
Why such a big scale? We just don't believe that a 5-Star or 10-Point system is enough to sufficiently understand a person's taste. Honestly, 100 points isn't really enough, either, but it's a start! With smaller scales, everything often ends up at "3.5" or "4 stars", and nuance is lost. That same clumping can happen even on a 100-point scale, but it's less pronounced. Criticker users with thousands of ratings know there's a big difference between a score of 75 and 76!
If you don't want to use the full scale, Criticker will adjust to whatever system you adopt... as long as there's at least some difference in your scores. Like, you can't rate everything either "1" or "0" and expect Criticker to understand your tastes.
2. Normalize ratings
Once you've got enough ratings, Criticker normalizes your scores. We do this because of the wide variety in the way people rate things. For some people, a score of "80" is close to a masterpiece, while for others "80" is middling. Some try to spread their scores equally across the whole scale, while others adhere to the system they learned in school, when a "60" was real bad. Like, if you came home with a "60" on your math test, you were probably going to get grounded.
Most often, ratings aren't evenly distributed
This is where percentiles come in. You rated Wrath of the Titans a "60", because you couldn't stand it. For you, that's a really bad score, and in fact is in the bottom 13% of all your ratings. However, for UserX, "60" means "not bad", while a score of "28" represents their 13th percentile. So if UserX gave Wrath of the Titans a "28"... you actually both agree. Despite the large gap between the scores themselves, you've both put it in your bottom 13th percentile of everything you've ever rated.
3. Match with others
Once you have at least 10 titles rated, Criticker will start matching you to other users. As long as you have 3 titles in common with another user, we will generate your Taste Compatibility Index (or TCI). This three-title minimum raises, the more ratings you have.
Let's continue the example with UserX. You both scored Wrath of the Titans in your 13th percentile... but it turns out you have a couple other films in common as well.
| You | UserX | Difference | |
|---|---|---|---|
| Wrath of the Titans | 60 (13th percentile) | 28 (13th percentile) | 0 |
| T2 Trainspotting | 72 (44th percentile) | 73 (74th percentile) | 30 |
| Poor Things | 95 (90th percentile) | 90 (92nd percentile) | 2 |
| Taste Compatibility Index | 10.67 | ||
You didn't quite see eye-to-eye on T2 Trainspotting, but still... a TCI of 10.67 is really good! In this same way, you'll build TCIs with other users at Criticker. We store your best 1000 matches (or 2000 for sponsors), and use their opinions to generate recommendations for you.
As part of Criticker's commitment toward total transparency, you can see the math behind any TCI calculation on each user's profile page. You'll see each film you have in common, along with the difference between your scores and the resulting TCI... or average of those differences.
4. Build recommendations
At this point, Criticker has compared your taste with thousands of other users, and found those who you match with most exactly. Now, it's time to build some recommendations.
As an example, when building a prediction for Tenet, Criticker gathers your top ten TCIs who've given it a rating. As long as at least three of your top 1000 TCIs have rated this film, a prediction can be made. The percentiles of their ratings are averaged, and that resulting average percentile is applied to your rating scale.
| Score | Percentile | |
|---|---|---|
| UserX | 90 | 92nd |
| UserY | 43 | 10th |
| UserX | 78 | 67th |
| Average Percentile | 56.33 | |
| Your PSI | 77 | |
On your rating scale, a percentile of 56.33 corresponds to a score of 77, which is your PSI for Tenet -- the rating Criticker predicts you'll give it. Not great, but not terrible either! This is how your recommendations are built, title by title, using the opinions of those users who most closely match your own tastes.
5. Keep things fresh
As you use Criticker, your TCIs and PSIs are being built behind the scenes. And whenever you visit a particular title, the PSI is generated from scratch. Over time, you'll build predictions for thousands, or even tens of thousands of titles, which you can then view in the PSI List, or the Title Database.
Of course, with every title you rate, your TCIs might change drastically -- especially at the beginning of your time on the site. And every time your TCIs change, your PSIs might change too. As Criticker learns more and more about your taste in movies, TV shows, and games, it never rests in calculating and recalculating the recommendations it provides to you.
And that's basically how Criticker works! If you have questions that weren't answered here, make sure to check out the Frequently Asked Questions, or get in touch with us! We always try to answer questions as soon as we're able.
Happy rating! We're sure you're going to find something amazing to watch, with Criticker's help!