How the Dot Product Helps You Find Love (The Math of Dating Apps)
HTML-код
- Опубликовано: 8 фев 2025
- In part 2 of our linear algebra series we cover the dot product and how it is used in matching algorithms like the cosine ranking method. These algorithms are widely used in matching services like dating apps and recommendations given to you on tv, movie, and music apps.
Math The World is dedicated to bringing real world math problems into the classroom and answering the age old question “when will I ever use this?”
Part 1: • Applied Linear Algebra...
Instagram: / maththeworld
Twitter/X: / math_the_world
Facebook: / maththeworldproject
Email: MathTheWorld@byu.edu
Created by Doug Corey
Script: Doug Corey and Jennifer Canizales
Audio: Doug Corey
Animation: Jennifer Canizales
Music: Coma Media \ Restum Anoush from Pixabay
© 2024 BYU
![ChatGPT is made from 100 million of these [The Perceptron]](http://i.ytimg.com/vi/l-9ALe3U-Fg/mqdefault.jpg)
![ChatGPT is made from 100 million of these [The Perceptron]](/img/tr.png)
funny how you translate math concepts into WTF examples xD
Thank you for making these videos, they really give the subjects some great extra intuition!
This is a cool video! I feel like cosine similarity is such a cool idea, I use it a lot in machine learning and data analysis stuff I do.
you'll love our next video then!
The Correlation between maths and dating is so profoundly explained by you ..
Amazing video! Gave me motivation to study for my linear exam, not by stressing me out with a bombardment of information but instead by showing the sheer power of on of the tools we’ve learned about.
I’m very excited for the next video!
I am hoping to be a data analyst in the future, but that involves getting some portfolio projects done. My friend has been doing these blind dating nights for two semesters now, and this semester we decided to create a mathematical model to match people appropriately. This video came at such a great time and we’re now using cosine similarity as a huge part of our project! What’s cool is that this could actually improve the blind date night, we’ve got about 160 submissions last semester.
Thanks for this clear explanation! Cool stuff!
You are welcome! I am glad you liked it!
That is so fun!! And allows such interesting experiments too
What would be a nice mathematical way to match opposites? Like if a student likes to do most of the work by themselves while another doesn't have much free time, so having opposite scores on this one would make a better match
i guess "opposite" is equivalent to "least similar", so dot product should still work for your use case, just multiply by -1
@bhargavganji6525 and could that work with mixed questions too? Having 5 questions that match when similar and having 5 questions that match when opposite
That sounds nice
i think a better method would be able to tell that, if we have a student A that said -1, a student B that said 1 and another student C that said 2, C is further from A than B. In the cosine method there is no difference between this two students, as Noon scored the minimum of -1, and if Noon had answered only -2 anf 2 depending on wich were further from Midnight, it theoretically should be a worse match. Also if Midnight said 0, Noon would say the same, so the are not really opposites.
Nice video though, good work
very informative packet yo!
glorious king
Paused at the beginning to think how I would do it. Thing is, without the result of dates you can't train the model so you have to assume similar is better.
So just a Gram matrix then. Or loop and find the minimum. It won't be very good. You're not biasing what's important, you're not allowing for anything other than linear concordance.
I would probably take the outer product and make the discrete cosine transform and then throw away the high frequency noise to avoid over fitting. Fundamentally though, you need some feedback to train.
Opposites Attract :D
in my physics major, the lineair algebra exam will really s*ck. But, Im sure I will get it (I was sick and had strugled in my life what became a trouble for my time). So, I had really late began to learn the lineair map topic. I regret it, but fine.
These video is cool, because these topic really matters (maybe are dating apps not so important as food or biotechnology, but however, the link with math is fascinating).
Title is misleading. The apps are not meant to find you love, they're meant to get you to pay for them, and as long as possible, while working just well enough that people will still use them. For example, there's generally several times as many men on apps as women. It is literally impossible to match everyone, even if preferences were weak. What actually happens is a fraction of the men get most of the right swipes/likes/etc, and the women get passed around by them and complain that all men suck. How in the heck can you match people that will likely marry with this input? It's like trying to make a cake out of a turd. All you end up doing is polishing the turd so people will pay to come look at it
I’m cooked
I get a lot of matches and I just don’t message them, if I do it gets past 3 messages and fizzles out
Dot product needs a weight filter
So.... I guess we are just assuming a good partner is someone that are like yourself? 😂
I love dot products, use the all the time. But when trying to find a partner I value "complimenting" just as high as "sameness".
And even for movie recommendations i don't think it is that good of a tool. Yes I love The Matrix and Ghost In The Shell (probably close in vector space of movies), but I also love LOTR, princess mononoke, Alien, Interstellar, Memento, Mr. Nobody, Pulp Fiction, Mad Max: Fury Road and The Accountant. Very different films that a probably located all over the place in vector space. What if I just want to see something that I have not seen yet, something different from what I already have seen.
I must say I am not super impressed by Netflix recommendations. I am honestly very confused how they can have all that data on me and still recommend me "paul vs tyson" and O.J Simpson. Half the time it just falls back on suggesting things I already have seen on Netflix. But I guess they are close in vector space 😂.
I remember reading somewhere that Napoleon Dynamite was such a difficult movie in recommendations because it doesn't really belong near any of the other movies in vector space.
The way dating apps apply cosine ranking and other matching algorithms that utilize the dot product is a little more complex and we'll be diving into that more in our next video! But in short they aren't directly matching you to the person you are swiping on. Instead they take people you have swiped on, match you to other people who have swiped on those same people as you, and then use them as a match to feed you more people you might swipe on. This is easier to explain with movies. If I like horror movies, they'll match me to someone who has also watched some of the horror movies I have and then use their watch list to feed me more horror movies I might like because the person I matched to also liked those movies. So in this example the people you swipe on in dating apps are like the horror movies, but the matching algorithm isn't directly matching you to them, they're matching you to other horror movie lovers.
UNBELIEVABLY OLD-FOR-AGE-LOOKING WOMEN
Yohoho