Inside OKCupid: The math of online dating - Christian Rudder

1,238,889 views ใƒป 2013-02-13

TED-Ed


ืื ื ืœื—ืฅ ืคืขืžื™ื™ื ืขืœ ื”ื›ืชื•ื‘ื™ื•ืช ื‘ืื ื’ืœื™ืช ืœืžื˜ื” ื›ื“ื™ ืœื”ืคืขื™ืœ ืืช ื”ืกืจื˜ื•ืŸ.

00:00
Translator: Andrea McDonough Reviewer: Bedirhan Cinar
0
0
7000
ืชืจื’ื•ื: Ido Dekkers ืขืจื™ื›ื”: Sigal Tifferet
00:17
Hello, my name is Christian Rudder,
1
17903
1714
ืฉืœื•ื, ืฉืžื™ ื›ืจื™ืกื˜ื™ืืŸ ืจื•ื“ืจ,
00:19
and I was one of the founders of OkCupid.
2
19641
2209
ื•ืื ื™ ื”ื™ื™ืชื™ ืื—ื“ ื”ืžืงื™ืžื™ื ืฉืœ OK ืงื•ืคื™ื“ื•ืŸ.
00:21
It's now one of the biggest dating sites in the United States.
3
21874
2918
ื›ืจื’ืข ื–ื” ืื—ื“ ืžืืชืจื™ ื”ื”ื›ืจื•ื™ื•ืช ื”ื’ื“ื•ืœื™ื ื‘ื™ื•ืชืจ ื‘ืืจืฆื•ืช ื”ื‘ืจื™ืช.
00:24
Like most everyone at the site, I was a math major,
4
24816
2391
ื›ืžื• ื›ืžืขื˜ ื›ื•ืœื ื‘ืืชืจ,
ื”ื™ื™ืชื™ ื‘ื•ื’ืจ ืžืชืžื˜ื™ืงื”, ื•ื›ืžื• ืฉืืชื ื™ื›ื•ืœื™ื ืœืฆืคื•ืช,
00:27
As you may expect, we're known for the analytic approach we take to love.
5
27231
3440
ืื ื—ื ื• ื™ื“ื•ืขื™ื ื‘ื’ื™ืฉื” ื”ืื ืœื™ื˜ื™ืช
ืฉื™ืฉ ืœื ื• ืœืื”ื‘ื”.
ืื ื—ื ื• ืงื•ืจืื™ื ืœื” ืืœื’ื•ืจื™ืชื ื”ื”ืชืืžื” ืฉืœื ื•.
00:30
We call it our matching algorithm.
6
30695
1638
ื‘ืขื™ืงืจื•ืŸ ืืœื’ื•ืจื™ืชื ื”ื”ืชืืžื” ืฉืœ OK ืงื•ืคื™ื“ื•ืŸ
00:32
Basically, OkCupid's matching algorithm helps us decide
7
32357
2588
ืขื•ื–ืจ ืœื ื• ืœื”ื—ืœื™ื˜ ืื ืฉื ื™ ืื ืฉื™ื ืฆืจื™ื›ื™ื ืœืฆืืช ืœืคื’ื™ืฉื”.
00:34
whether two people should go on a date.
8
34969
1876
00:36
We built our entire business around it.
9
36869
1872
ื‘ื ื™ื ื• ืืช ื›ืœ ื”ืขืกืง ืกื‘ื™ื‘ ื–ื”.
00:38
Now, algorithm is a fancy word,
10
38765
1960
ืขื›ืฉื™ื• ืืœื’ื•ืจื™ืชื ื–ื• ืžื™ืœื” ืžืคื•ืืจืช,
00:40
and people like to drop it like it's this big thing.
11
40749
2485
ื•ืื ืฉื™ื ืื•ื”ื‘ื™ื ืœื”ื’ื™ื“ ืื•ืชื” ื›ืื™ืœื• ื–ื” ื“ื‘ืจ ื’ื“ื•ืœ ื›ื–ื”,
00:43
But really, an algorithm is just a systematic,
12
43258
2288
ืื‘ืœ, ืœืžืขืฉื”, ืืœื’ื•ืจื™ืชื ื”ื•ื ื“ืจืš,
00:45
step-by-step way to solve a problem.
13
45570
2223
ืฉื™ื˜ืชื™ืช ืฉืœ ืฉืœื‘ ืื—ืจ ืฉืœื‘ ืœืคืชื•ืจ ื‘ืขื™ื”.
00:47
It doesn't have to be fancy at all.
14
47817
2177
ื–ื” ืœื ื—ื™ื™ื‘ ืœื”ื™ื•ืช ืžืคื•ืืจ ื‘ื›ืœืœ.
ื›ืืŸ, ื‘ืฉื™ืขื•ืจ ื”ื–ื”, ืื ื™ ืขื•ืžื“ ืœื”ืกื‘ื™ืจ
00:50
Here in this lesson,
15
50018
1151
00:51
I'm going to explain how we arrived at our particular algorithm,
16
51193
3008
ืื™ืš ื”ื’ืขื ื• ืœืืœื’ื•ืจื™ืชื ื”ื™ื—ื•ื“ื™ ืฉืœื ื•
ื›ืš ืฉืชืจืื• ืื™ืš ื–ื” ื ืขืฉื”.
00:54
so you can see how it's done.
17
54225
1411
00:55
Now, why are algorithms even important?
18
55660
1934
ืขื›ืฉื™ื•, ืœืžื” ืืœื’ื•ืจื™ืชืžื™ื ื‘ื›ืœืœ ื—ืฉื•ื‘ื™ื?
00:57
Why does this lesson even exist?
19
57618
1580
ืœืžื” ื”ืฉื™ืขื•ืจ ื”ื–ื” ื‘ื›ืœืœ ืงื™ื™ื?
00:59
Well, notice one very significant phrase I used above:
20
59222
3420
ื•ื‘ื›ืŸ, ืฉื™ืžื• ืœื‘ ืœื‘ื™ื˜ื•ื™ ืื—ื“ ื—ืฉื•ื‘ ืœืžืขืœื” ืฉื”ืฉืชืžืฉืชื™ ื‘ื•:
01:02
they are a step-by-step way to solve a problem,
21
62666
2339
ื™ืฉ ื“ืจืš ืฉืœ ืฉืœื‘ ืื—ืจ ืฉืœื‘ ืœืคืชื•ืจ ืืช ื”ื‘ืขื™ื”,
01:05
and as you probably know, computers excel at step-by-step processes.
22
65029
3418
ื•ื›ืžื• ืฉืืชื ื‘ื•ื•ื“ืื™ ื™ื•ื“ืขื™ื,
ืžื—ืฉื‘ื™ื ืžืฆื˜ื™ื™ื ื™ื ื‘ืชื”ืœื™ื›ื™ื ืฉืœ ืฉืœื‘ ืื—ืจ ืฉืœื‘.
01:08
A computer without an algorithm
23
68471
1589
ืžื—ืฉื‘ ื‘ืœื™ ืืœื’ื•ืจื™ืชื
ื”ื•ื ื‘ืขื™ืงืจื•ืŸ ืžืฉืงื•ืœืช ื ื™ื™ืจ ื™ืงืจื”.
01:10
is basically an expensive paperweight.
24
70084
2724
01:12
And since computers are such a pervasive part of everyday life,
25
72832
2989
ื•ืžืื—ืจ ื•ืžื—ืฉื‘ื™ื ื”ื ื“ื‘ืจ ื›ื” ื ืคื•ืฅ ื‘ื—ื™ื™ื ื”ื™ื•ื ื™ื•ืžื™ื™ื,
01:15
algorithms are everywhere.
26
75845
1547
ืืœื’ื•ืจื™ืชืžื™ื ื”ื ื‘ื›ืœ ืžืงื•ื.
01:18
The math behind OkCupid's matching algorithm is surprisingly simple.
27
78590
3197
ื”ืžืชืžื˜ื™ืงื” ืžืื—ื•ืจื™ ืืœื’ื•ืจื™ืชื ื”ื”ืชืืžื” ืฉืœ OK ืงื•ืคื™ื“ื•ืŸ
ื”ื™ื ืคืฉื•ื˜ื” ืœื”ืคืœื™ื.
01:21
It's just some addition, multiplication, a little bit of square roots.
28
81811
4002
ื–ื” ืคืฉื•ื˜ ืงืฆืช ื—ื™ื‘ื•ืจ,
ื›ืคืœ,
ื•ืžืขื˜ ืฉื•ืจืฉื™ื ืžืจื•ื‘ืขื™ื.
01:25
The tricky part in designing it
29
85837
1690
ื”ื—ืœืง ื”ืงืฉื” ื‘ืœืชื›ื ืŸ ืืช ื–ื”, ืขื ื–ืืช,
01:27
was figuring out how to take something mysterious,
30
87551
2565
ื”ื™ื” ืœื”ื‘ื™ืŸ ืื™ืš ืœืงื—ืช ืžืฉื”ื• ืžืกืชื•ืจื™,
01:30
human attraction,
31
90140
1150
ืžืฉื™ื›ื” ืื ื•ืฉื™ืช,
01:31
and break it into components that a computer can work with.
32
91314
2784
ื•ืœืคืจืง ืืช ื–ื” ืœื—ืœืงื™ื ืฉืžื—ืฉื‘ ื™ื›ื•ืœ ืœืขื‘ื•ื“ ืื™ืชื.
ื•ื‘ื›ืŸ, ื”ื“ื‘ืจ ื”ืจืืฉื•ืŸ ืฉื”ื™ื™ื ื• ืฆืจื™ื›ื™ื ื›ื“ื™ ืœืฉื“ืš ืื ืฉื™ื ื”ื™ื” ืžื™ื“ืข,
01:34
The first thing we needed to match people up was data,
33
94122
2553
01:36
something for the algorithm to work with.
34
96699
1992
ืžืฉื”ื• ืœืืœื’ื•ืจื™ืชื ืœืขื‘ื•ื“ ืขืœื™ื•.
01:38
The best way to get data quickly from people is to just ask for it.
35
98715
3158
ื”ื“ืจืš ื”ื˜ื•ื‘ื” ื‘ื™ื•ืชืจ ื›ื“ื™ ืœื”ืฉื™ื’ ืžื™ื“ืข ืžืื ืฉื™ื ื‘ืžื”ื™ืจื•ืช
ื”ื•ื ืคืฉื•ื˜ ืœื‘ืงืฉ ืื•ืชื•.
01:41
So we decided that OkCupid should ask users questions,
36
101897
2727
ืื–, ื”ื—ืœื˜ื ื• ืฉOK ืงื•ืคื™ื“ื•ืŸ ื™ืฉืืœ ืื ืฉื™ื ืฉืืœื•ืช,
01:44
stuff like, "Do you want to have kids one day?"
37
104648
2357
ื“ื‘ืจื™ื ื›ืžื•, "ืืชื” ืจื•ืฆื” ื™ืœื“ื™ื ื™ื•ื ืื—ื“?"
ื• "ื‘ืื™ื–ื• ืชื›ื™ืคื•ืช ืืชื” ืžืฆื—ืฆื— ืืช ื”ืฉื™ื ื™ื™ื?",
01:47
"How often do you brush your teeth?"
38
107029
1758
01:48
"Do you like scary movies?"
39
108811
1392
"ืืชื” ืื•ื”ื‘ ืกืจื˜ื™ื ืžืคื—ื™ื“ื™ื?"
01:50
And big stuff like, "Do you believe in God?"
40
110675
2077
ื•ื“ื‘ืจื™ื ื’ื“ื•ืœื™ื ื›ืžื• "ืืชื” ืžืืžื™ืŸ ื‘ืืœื•ื”ื™ื?"
01:53
Now, a lot of the questions are good for matching like with like,
41
113843
3064
ืขื›ืฉื™ื•, ื”ืจื‘ื” ืžื”ืฉืืœื•ืช ื˜ื•ื‘ื•ืช
ืœื”ืชืืžืช ืชื—ื•ืžื™ ืขื ื™ื™ืŸ,
01:56
that is, when both people answer the same way.
42
116931
2156
ื–ื” ื›ืฉืฉื ื™ ื”ืื ืฉื™ื ืขื•ื ื™ื ืื•ืชื• ื”ื“ื‘ืจ.
01:59
For example, two people who are both into scary movies
43
119111
2548
ืœื“ื•ื’ืžื”, ืฉื ื™ ืื ืฉื™ื ืฉืื•ื”ื‘ื™ื ืกืจื˜ื™ื ืžืคื—ื™ื“ื™ื
02:01
are probably a better match than one person who is and one who isn't.
44
121683
3321
ื”ื ื›ื ืจืื” ื”ืชืืžื” ื˜ื•ื‘ื” ื™ื•ืชืจ
ืžืื“ื ืื—ื“ ืฉืื•ื”ื‘
ื•ืื“ื ืฉื ื™ ืฉืœื.
02:05
But what about a question like,
45
125028
1493
ืื‘ืœ ืžื” ืขื ืฉืืœื•ืช ื›ืžื•,
02:06
"Do you like to be the center of attention?"
46
126545
2062
"ืืชื” ืื•ื”ื‘ ืœื”ื™ื•ืช ื‘ืžืจื›ื– ื”ืขื ื™ื™ื ื™ื?"
02:08
If both people in a relationship are saying yes to this,
47
128631
2628
ืื ืฉื ื™ ื”ืื ืฉื™ื ื‘ื™ื—ืกื™ื ืื•ืžืจื™ื ื›ืŸ ืœื–ื”,
ืื– ื”ื ื™ื”ื™ื• ื‘ื‘ืขื™ื” ื’ื“ื•ืœื”.
02:11
they're going to have massive problems.
48
131283
2093
02:13
We realized this early on,
49
133400
1245
ื”ื‘ื ื• ืืช ื–ื” ื“ื™ ื‘ื”ืชื—ืœื”,
02:14
and so we decided we needed a bit more data from each question.
50
134669
3269
ื•ืื– ื”ื—ืœื˜ื ื• ืฉืื ื—ื ื• ืฆืจื™ื›ื™ื
ื™ื•ืชืจ ืžื™ื“ืข ืžื›ืœ ืฉืืœื”.
02:17
We had to ask people to specify not only their own answer,
51
137962
2763
ื”ื™ื™ื ื• ืฆืจื™ื›ื™ื ืœื‘ืงืฉ ืžืื ืฉื™ื ืœืคืจื˜ ืœื ืจืง ืืช ื”ืชืฉื•ื‘ื” ืฉืœื”ื,
02:20
but the answer they wanted from someone else.
52
140749
2265
ืืœื ืืช ื”ืชืฉื•ื‘ื” ืฉื”ื ืจืฆื• ืžื”ืื“ื ื”ืฉื ื™.
02:23
That worked really well.
53
143038
1501
ื–ื” ืขื‘ื“ ืžืžืฉ ื˜ื•ื‘,
02:24
But we needed one more dimension.
54
144563
1604
ืื‘ืœ ื”ื™ื™ื ื• ืฆืจื™ื›ื™ื ืขื•ื“ ืžื™ืžื“.
02:26
Some questions tell you more about a person than others.
55
146191
2643
ื›ืžื” ืžื”ืฉืืœื•ืช ืžืกืคืจื•ืช ืœื›ื ืขืœ ื”ืื“ื ื™ื•ืชืจ ืžืื—ืจื•ืช.
02:28
For example, a question about politics, something like,
56
148858
3395
ืœื“ื•ื’ืžื”, ืฉืืœื” ืขืœ ืคื•ืœื™ื˜ื™ืงื”, ืžืฉื”ื• ื›ืžื•,
"ืžื” ื’ืจื•ืข ื™ื•ืชืจ: ืฉืจื™ืคืช ืกืคืจื™ื ืื• ืฉืจื™ืคืช ื“ื’ืœื™ื?"
02:32
"Which is worse: book burning or flag burning?"
57
152277
2288
02:34
might reveal more about someone than their taste in movies.
58
154589
2810
ืื•ืœื™ ืชื’ืœื” ื™ื•ืชืจ ืขืœ ืžื™ืฉื”ื• ืžื”ื˜ืขื ืฉืœื”ื ื‘ืกืจื˜ื™ื.
02:37
And it doesn't make sense to weigh all things equally,
59
157423
2619
ื•ื–ื” ืœื ื”ื’ื™ื•ื ื™ ืœืฉืงืœืœ ืืช ื›ืœ ื”ื“ื‘ืจื™ื ื‘ืžืฉืงืœ ื–ื”ื”,
ืื– ืื ื—ื ื• ืฆืจื™ื›ื™ื ืœื”ื•ืกื™ืฃ ืขื•ื“ ื ืงื•ื“ืช ืžื™ื“ืข ืื—ืจื•ื ื”.
02:40
so we added one final data point.
60
160066
1596
02:41
For everything that OkCupid asks you,
61
161686
2024
ืœื›ืœ ื“ื‘ืจ ืฉOK ืงื•ืคื™ื“ื•ืŸ ืฉื•ืืœ ืืชื›ื,
02:43
you have a chance to tell us the role it plays in your life.
62
163734
2829
ื™ืฉ ืœื›ื ืืคืฉืจื•ืช ืœื”ื’ื™ื“ ืœื ื•
ืืช ื”ืชืคืงื™ื“ ืฉื–ื” ืžืฉื—ืง ื‘ื—ื™ื™ื›ื,
02:46
And this ranges from irrelevant to mandatory.
63
166587
2319
ื•ื–ื” ื ืข ืžืœื ืจืœื•ื•ื ื˜ื™ ืœื”ื›ืจื—ื™.
02:49
So now, for every question, we have three things for our algorithm:
64
169446
3222
ืื– ืขื›ืฉื™ื•, ืœื›ืœ ืฉืืœื”,
ื™ืฉ ืœื ื• ืฉืœื•ืฉื” ื“ื‘ืจื™ื ืœืืœื’ื•ืจื™ืชื ืฉืœื ื•:
02:52
first, your answer;
65
172692
1352
ืจืืฉื™ืช, ืืช ื”ืชืฉื•ื‘ื” ืฉืœื›ื;
02:54
second, how you want someone else -- your potential match -- to answer;
66
174617
4140
ืฉื ื™ืช, ืื™ืš ื”ื™ื™ืชื ืจื•ืฆื™ื ืฉืžื™ืฉื”ื• ืื—ืจ,
ื”ื”ืชืืžื” ื”ืคื•ื˜ื ืฆื™ืืœื™ืช ืฉืœื›ื,
ื™ืขื ื”;
02:58
and third, how important the question is to you at all.
67
178781
2788
ื•ืฉืœื™ืฉื™ืช, ื›ืžื” ื”ืฉืืœื” ื‘ื›ืœืœ ื—ืฉื•ื‘ื” ืœื›ื.
03:02
With all this information,
68
182710
1252
ืขื ื›ืœ ื”ืžื™ื“ืข ื”ื–ื”,
03:03
OkCupid can figure out how well two people will get along.
69
183986
3118
OK ืงื•ืคื™ื“ื•ืŸ ื™ื›ื•ืœ ืœื”ื‘ื™ืŸ ื›ืžื” ืฉื ื™ ืื ืฉื™ื ื™ื›ื•ืœื™ื ืœื”ืชืื™ื.
03:07
The algorithm crunches the numbers and gives us a result.
70
187128
3006
ื”ืืœื’ื•ืจื™ืชื ืžืขื‘ื“ ืืช ื”ืžืกืคืจื™ื ื•ื ื•ืชืŸ ืœื ื• ืชื•ืฆืื”.
ื›ื“ื•ื’ืžื” ืžืขืฉื™ืช,
03:10
As a practical example,
71
190158
1152
03:11
let's look at how we'd match you with another person.
72
191334
2525
ื‘ื•ืื• ื ืจืื” ืื™ืš ื ืชืื™ื ืืชื›ื ืœืื“ื ืื—ืจ,
03:13
Let's call him "B."
73
193883
1189
ื‘ื•ืื• ื ืงืจื ืœื•, "ื‘".
ื”ืชืืžืช ื”ืื—ื•ื–ื™ื ืฉืœื›ื ืขื ื‘ ืžื‘ื•ืกืกืช ืขืœ
03:16
Your match percentage with B is based on questions you've both answered.
74
196023
3482
ืฉืืœื•ืช ืฉืฉื ื™ื›ื ืขื ื™ืชื.
03:19
Let's call that set of common questions "s."
75
199529
2425
ื‘ื•ืื• ื ืงืจื ืœืกื˜ ื”ื–ื” ืฉืœ ืฉืืœื•ืช ืžืฉื•ืชืคื•ืช, "ืก".
ื›ื“ื•ื’ืžื” ืžืžืฉ ืคืฉื•ื˜ื”, ื ืฉืชืžืฉ ื‘ืกื˜ ืงื˜ืŸ "ืก"
03:22
As a very simple example, we use a small set "s"
76
202559
2349
03:24
with just two questions in common,
77
204932
1641
ืขื ืจืง ืฉืชื™ ืฉืืœื•ืช ืžืฉื•ืชืคื•ืช
03:26
and compute a match from that.
78
206597
1828
ื•ื ื—ืฉื‘ ื”ืชืืžื” ืœืคื™ ื–ื”.
03:28
Here are our two example questions.
79
208449
1671
ื”ื ื” ืฉืชื™ ืฉืืœื•ืช ื”ื“ื•ื’ืžื” ืฉืœื ื•.
03:30
The first one, let's say, is, "How messy are you?"
80
210144
2381
ื”ืจืืฉื•ื ื”, ื ื’ื™ื“, ื”ื™ื," ื›ืžื” ืžื‘ื•ืœื’ืŸ ืืชื”?"
03:32
And the answer possibilities are:
81
212549
2096
ื•ื”ืชืฉื•ื‘ื•ืช ื”ืืคืฉืจื™ื•ืช ื”ืŸ
03:34
very messy, average and very organized.
82
214669
3361
ืžืื•ื“ ืžื‘ื•ืœื’ืŸ,
ืžืžื•ืฆืข,
ื•ืžืื•ื“ ืžืกื•ื“ืจ.
03:38
And let's say you answered "very organized,"
83
218054
2060
ื•ื‘ื•ืื• ื ื’ื™ื“ ืฉืขื ื™ืชื "ืžืื•ื“ ืžืกื•ื“ืจ,"
ื•ื”ื™ื™ืชื ืจื•ืฆื™ื ืžื™ืฉื”ื• ืื—ืจ ืฉืขื ื” "ืžืื•ื“ ืžืกื•ื“ืจื™ื,"
03:40
and you'd like someone else to answer "very organized,"
84
220138
2760
03:42
and the question is very important to you.
85
222922
2256
ื•ื”ืฉืืœื” ื”ื™ื ืžืื•ื“ ื—ืฉื•ื‘ื” ืœื›ื.
03:45
Basically, you're a neat freak.
86
225202
1492
ื‘ืขื™ืงืจื•ืŸ ืืชื ืžืฉื•ื’ืขื™ื ืœืกื“ืจ.
03:46
You're neat, you want someone else to be neat, and that's it.
87
226718
2868
ืืชื ืžืกื•ื“ืจื™ื,
ืืชื ืจื•ืฆื™ื ืžื™ืฉื”ื• ืื—ืจ ืฉื™ื”ื™ื” ืžืกื•ื“ืจ,
ื•ื–ื”ื• ื–ื”.
03:49
And let's say B is a little bit different.
88
229610
2015
ื•ื‘ื•ืื• ื ื’ื™ื“ ืฉ"ื‘" ื”ื•ื ืžืขื˜ ืฉื•ื ื”.
03:51
He answered "very organized" for himself,
89
231649
2039
ื”ื•ื ืขื ื” ืžืื•ื“ ืžืกื•ื“ืจ ืขืœ ืขืฆืžื•,
03:53
but "average" is OK with him as an answer from someone else,
90
233712
3007
ืื‘ืœ ืžืžื•ืฆืข ื‘ืกื“ืจ ืœื•
ื›ืชืฉื•ื‘ื” ืฉืœ ืžื™ืฉื”ื• ืื—ืจ,
03:56
and the question is only a little important to him.
91
236743
2402
ื•ื”ืฉืืœื” ืจืง ืžืขื˜ ื—ืฉื•ื‘ื” ืœื•.
ื‘ื•ืื• ื ื‘ื™ื˜ ื‘ืฉืืœื” ื”ืฉื ื™ื”,
03:59
Let's look at the second question, from our previous example:
92
239169
2893
ื”ื™ื ื–ืืช ืžื”ื“ื•ื’ืžื” ื”ืงื•ื“ืžืช ืฉืœื ื•:
"ื”ืื ืืชื ืื•ื”ื‘ื™ื ืœื”ื™ื•ืช ืžืจื›ื– ื”ืขื ื™ื™ื ื™ื?"
04:02
"Do you like to be the center of attention?"
93
242086
2056
ื”ืชืฉื•ื‘ื•ืช ื”ืŸ ืจืง ื›ืŸ ื•ืœื.
04:04
The answers are "yes" and "no."
94
244166
1514
04:05
You've answered "no," you want someone else to answer "no,"
95
245704
2995
ืขื›ืฉื™ื• ืืชื ืขื ื™ืชื "ืœื,"
ื•ืจืฆื™ืชื ืฉื’ื ื”ืฉื ื™ ื™ืขื ื” "ืœื,"
04:08
and the question is only a little important to you.
96
248723
2391
ื•ื”ืฉืืœื” ื”ื™ื ืจืง ืžืขื˜ ื—ืฉื•ื‘ื” ืœื›ื.
ืขื›ืฉื™ื• "ื‘", ืขื ื” "ื›ืŸ,"
04:11
Now B, he's answered "yes."
97
251138
1621
04:12
He wants someone else to answer "no,"
98
252783
1776
ื•ื”ื•ื ืจื•ืฆื” ืฉื”ืื—ืจ ื™ืขื ื” "ืœื,"
04:14
because he wants the spotlight on him,
99
254583
2274
ืžืคื ื™ ืฉื”ื•ื ืจื•ืฆื” ืืช ืื•ืจ ื”ื–ืจืงื•ืจื™ื ืขืœื™ื•,
04:16
and the question is somewhat important to him.
100
256881
2430
ื•ื”ืฉืืœื” ื”ื™ื ืžืขื˜ ื—ืฉื•ื‘ื” ืœื•.
04:19
So, let's try to compute all of this.
101
259335
1999
ืื–, ื‘ื•ืื• ื ื ืกื” ืœื—ืฉื‘ ืืช ื›ืœ ื–ื”.
04:21
Our first step is, since we use computers to do this,
102
261972
2503
ื”ืฉืœื‘ ื”ืจืืฉื•ืŸ ืฉืœื ื• ื”ื•ื,
ืžืื—ืจ ื•ืื ื—ื ื• ืžืฉืชืžืฉื™ื ื‘ืžื—ืฉื‘ื™ื ื›ื“ื™ ืœืขืฉื•ืช ืืช ื–ื”,
04:24
we need to assign numerical values
103
264499
1867
ืื ื—ื ื• ืฆืจื™ื›ื™ื ืœืฉื™ื™ืš ืขืจื›ื™ื ืžืกืคืจื™ื™ื
04:26
to ideas like "somewhat important" and "very important,"
104
266390
2627
ืœืจืขื™ื•ื ื•ืช ื›ืžื• "ืžืขื˜ ื—ืฉื•ื‘" ื•ืžืื•ื“ ื—ืฉื•ื‘"
04:29
because computers need everything in numbers.
105
269041
2211
ืžืคื ื™ ืฉืžื—ืฉื‘ื™ื ืฆืจื™ื›ื™ื ื”ื›ืœ ื‘ืžืกืคืจื™ื.
04:31
We at OkCupid decided on the following scale:
106
271276
2403
ืื ื—ื ื• ื‘ OK ืงื•ืคื™ื“ื•ืŸ ื”ื—ืœื˜ื ื• ืขืœ ื”ืžื“ื“ ื”ื‘ื:
04:33
"Irrelevant" is worth 0.
107
273703
1946
ืœื ืจืœื•ื•ื ื˜ื™ ืฉื•ื•ื” 0,
ืงืฆืช ื—ืฉื•ื‘ ื–ื” 1,
04:36
"A little important" is worth 1.
108
276173
1889
04:38
"Somewhat important" is worth 10.
109
278538
1809
ื“ื™ ื—ืฉื•ื‘ ืฉื•ื•ื” 10,
04:40
"Very important" is 50.
110
280831
1754
ืžืื•ื“ ื—ืฉื•ื‘ ื–ื” 50,
04:42
And "absolutely mandatory" is 250.
111
282609
3612
ื•ื”ื›ืจื—ื™ ืœื—ืœื•ื˜ื™ืŸ ื–ื” 250.
04:46
Next, the algorithm makes two simple calculations.
112
286245
2631
ืื—ืจื™ ื–ื”, ื”ืืœื’ื•ืจื™ืชื ืขื•ืฉื” ืฉื ื™ ื—ื™ืฉื•ื‘ื™ื ืคืฉื•ื˜ื™ื.
04:48
The first is: How much did B's answers satisfy you?
113
288900
3246
ื”ืจืืฉื•ืŸ ื”ื•ื ื›ืžื” ื”ืชืฉื•ื‘ื•ืช ืฉืœ "ื‘" ืžืกืคืงื•ืช ืืชื›ื,
ืฉื–ื” ืื•ืžืจ, ื›ืžื” ื ืงื•ื“ื•ืช ืืคืฉืจื™ื•ืช "ื‘" ืงื™ื‘ืœ ื‘ืžื“ื“ ืฉืœื›ื?
04:52
That is, how many possible points did B score on your scale?
114
292170
3793
04:55
Well, you indicated that B's answer to the first question,
115
295987
3212
ื•ื‘ื›ืŸ, ืืžืจืชื ืฉื”ืชืฉื•ื‘ื” ืฉืœ "ื‘"
ืœืฉืืœื” ื”ืจืืฉื•ื ื” ืขืœ ืกื“ืจ
04:59
about messiness,
116
299223
1166
ื”ื™ื ืžืื•ื“ ื—ืฉื•ื‘ื” ืœื›ื.
05:00
was very important to you.
117
300413
1350
05:01
It's worth 50 points and B got that right.
118
301787
2230
ื”ื™ื ืฉื•ื•ื” 50 ื ืงื•ื“ื•ืช ื•"ื‘" ืงืœืข ืืœื™ื”.
05:04
The second question is worth only 1,
119
304375
1737
ื”ืฉืืœื” ื”ืฉื ื™ื” ืฉื•ื•ื” ืจืง 1
ืžืคื ื™ ืฉืืžืจืชื ืฉื–ื” ืจืง ืงืฆืช ื—ืฉื•ื‘ ืœื›ื,
05:06
because you said it was only a little important.
120
306136
2278
ื•"ื‘" ืœื ืงืœืข ืœื–ื”.
05:08
B got that wrong,
121
308438
1197
05:09
so B's answers were 50 out of 51 possible points.
122
309659
2782
ืื– ื”ืชืฉื•ื‘ื•ืช ืฉืœ "ื‘" ื”ื™ื• 50 ืžืชื•ืš 51 ื ืงื•ื“ื•ืช ืืคืฉืจื™ื•ืช.
05:12
That's 98% satisfactory. Pretty good.
123
312465
2608
ื–ื” ืกื™ืคื•ืง ืฉืœ 98%.
ื–ื” ื“ื™ ื˜ื•ื‘.
05:15
The second question the algorithm looks at is: How much did you satisfy B?
124
315097
3949
ื•ื”ืฉืืœื” ื”ืฉื ื™ื” ืฉื”ืืœื’ื•ืจืชื ื‘ื•ื“ืง
ื–ื” ื›ืžื” ืืชื ืžืกืคืงื™ื ืืช "ื‘".
ื•ื‘ื›ืŸ, "ื‘" ื ืชืŸ ื ืงื•ื“ื” ืื—ืช ืœืชืฉื•ื‘ื” ืฉืœื›ื
05:19
Well, B placed 1 point on your answer to the messiness question
125
319070
3259
ืœืฉืืœืช ื”ืกื“ืจ
05:22
and 10 on your answer to the second.
126
322353
1953
ื• 10 ืขืœ ื”ืชืฉื•ื‘ื” ืฉืœื›ื ืœืฉื ื™ื”.
05:24
Of those 11, that's 1 plus 10, you earned 10 --
127
324745
3387
ืžืืœื”, 11, ื–ื” 1 ื•ืขื•ื“ 10,
ืืชื ื”ืจื•ื•ื—ืชื 10,
05:28
you guys satisfied each other on the second question.
128
328156
2595
ืืชื ืกื™ืคืงืชื ืื—ื“ ืืช ื”ืฉื ื™ ื‘ืฉืืœื” ื”ืฉื ื™ื”.
05:30
So your answers were 10 out of 11 equals 91 percent satisfactory to B.
129
330775
4242
ืื– ื”ืชืฉื•ื‘ื” ืฉืœื›ื ื”ื™ืชื” 10 ืžืชื•ืš 11
ืฉื–ื” ืฉื•ื•ื” ืœ 91% ืกื™ืคื•ืง ืœ "ื‘".
05:35
That's not bad.
130
335041
1151
ื–ื” ืœื ืจืข.
05:36
The final step is to take these two match percentages
131
336216
2507
ื”ืฉืœื‘ ื”ืื—ืจื•ืŸ ื”ื•ื ืœืงื—ืช ืืช ืฉืชื™ ื”ืชืืžื•ืช ื”ืื—ื•ื– ื”ืืœื•
05:38
and get one number for the both of you.
132
338747
1866
ื•ืœืงื‘ืœ ืžืกืคืจ ืื—ื“ ืœืฉื ื™ื›ื.
05:40
To do this, the algorithm multiplies your scores,
133
340637
2611
ื›ื“ื™ ืœืขืฉื•ืช ืืช ื–ื”, ื”ืืœื’ื•ืจื™ืชื ืžื›ืคื™ืœ ืืช ื”ืชื•ืฆืื•ืช ืฉืœื›ื,
ืื– ื”ื•ื ืœื•ืงื— ืืช ื”ืฉื•ืจืฉ ื”"n",
05:43
then takes the nth root,
134
343272
1665
05:44
where "n" is the number of questions.
135
344961
2183
ื›ืฉ "n" ื”ื•ื ืžืกืคืจ ื”ืฉืืœื•ืช.
ืžืคื ื™ ืฉ"ืก", ืฉื–ื” ืžืกืคืจ ื”ืฉืืœื•ืช,
05:47
Because s, which is the number of questions in this sample,
136
347168
2830
ื‘ื“ื•ื’ืžื” ื”ื–ื•, ื”ื•ื ืจืง 2,
05:50
is only 2,
137
350022
1841
05:51
we have: match percentage equals the square root
138
351887
3665
ื™ืฉ ืœื ื• ืื—ื•ื– ื”ืชืืžื” ืฉืฉื•ื•ื”
ืœืฉื•ืจืฉ ืจื™ื‘ื•ืขื™ ืฉืœ 98% ื›ืคื•ืœ 91%.
05:55
of 98 percent times 91 percent.
139
355576
2896
05:58
That equals 94 percent.
140
358496
1784
ื–ื” ืฉื•ื•ื” 94%.
06:00
That 94 percent is your match percentage with B.
141
360304
3204
ื”94% ื”ืืœื” ื”ื ืื—ื•ื– ื”ื”ืชืืžื” ืฉืœื›ื ืœ"ื‘".
06:03
It's a mathematical expression of how happy you'd be with each other,
142
363532
3243
ื–ื” ื‘ื™ื˜ื•ื™ ืžืชืžื˜ื™
ืฉืœ ื›ืžื” ืžืื•ืฉืจื™ื ืืชื ืชื”ื™ื• ืื—ื“ ืขื ื”ืฉื ื™
06:06
based on what we know.
143
366799
1183
ื‘ื”ืชื‘ืกืก ืขืœ ืžื” ืฉืื ื—ื ื• ื™ื•ื“ืขื™ื.
ืขื›ืฉื™ื•, ืœืžื” ื”ืืœื’ื•ืจื™ืชื ืžื›ืคื™ืœ ื•ืœื
06:08
Now, why does the algorithm multiply,
144
368006
1786
06:09
as opposed to, say, average the two match scores together,
145
369816
2769
ืžืžืฆืข ืืช ืฉืชื™ ื”ืชื•ืฆืื•ืช ื™ื—ื“
06:12
and do the square-root business?
146
372609
1670
ื•ืžื—ืฉื‘ ืฉื•ืจืฉ ืจื™ื‘ื•ืขื™?
06:14
In general, this formula is called the geometric mean.
147
374303
2529
ื‘ื›ืœืœื™, ื”ื ื•ืกื—ื” ื ืงืจืืช ื”ืžืžื•ืฆืข ื”ื’ืื•ืžื˜ืจื™,
06:16
It's a great way to combine values that have wide ranges
148
376856
2627
ืฉื–ื• ื“ืจืš ืžืขื•ืœื” ืœืฉืœื‘ ืขืจื›ื™ื
ืฉื™ืฉ ืœื”ื ื˜ื•ื•ื— ืจื—ื‘
06:19
and represent very different properties.
149
379507
1915
ื•ืžื™ื™ืฆื’ื™ื ืชื›ื•ื ื•ืช ืฉื•ื ื•ืช ืžืื•ื“.
ื‘ืžื™ืœื™ื ืื—ืจื•ืช, ื–ื” ืžื•ืฉืœื ืœื”ืชืืžื” ืจื•ืžื ื˜ื™ืช.
06:21
In other words, it's perfect for romantic matching.
150
381446
2413
06:23
You've got wide ranges and you've got tons of different data points,
151
383883
3247
ื™ืฉ ืœื›ื ื˜ื•ื•ื— ืจื—ื‘
ื•ื™ืฉ ืœื›ื ื”ืžื•ืŸ ื ืงื•ื“ื•ืช ืžื™ื“ืข,
ื›ืžื• ืฉืืžืจืชื™, ืขืœ ืกืจื˜ื™ื,
06:27
like I said, about movies, politics, religion -- everything.
152
387154
3438
ืขืœ ืคื•ืœื™ื˜ื™ืงื”,
ืขืœ ื“ืช,
ืขืœ ื”ื›ืœ.
06:30
Intuitively, too, this makes sense.
153
390616
1838
ื‘ืื•ืคืŸ ืื™ื ื˜ื•ืื™ื˜ื™ื‘ื™ ื–ื” ื”ื’ื™ื•ื ื™.
06:32
Two people satisfying each other 50 percent
154
392478
2775
ืฉื ื™ ืื ืฉื™ื ืฉืžืกืคืงื™ื ืื—ื“ ืืช ื”ืฉื ื™ 50%
ืฆืจื™ื›ื™ื ืœื”ื™ื•ืช ื”ืชืืžื” ื˜ื•ื‘ื” ื™ื•ืชืจ
06:35
should be a better match than two others who satisfy 0 and 100,
155
395277
3952
ืžืืœื” ืฉืžืกืคืงื™ื ืื—ื“ ืืช ื”ืฉื ื™ 0 ื• 100,
06:39
because affection needs to be mutual.
156
399253
1814
ืžืคื ื™ ืฉื—ื™ื‘ื” ืฆืจื™ื›ื” ืœื”ื™ื•ืช ื”ื“ื“ื™ืช.
ืื—ืจื™ ื”ื•ืกืคืช ืชื™ืงื•ืŸ ืงื˜ืŸ ืœืžืจื•ื•ื— ื˜ืขื•ืช,
06:41
After adding a little correction for margin of error,
157
401091
2491
06:43
in the case where we have a small number of questions,
158
403606
2571
ื‘ืžืงืจื” ืฉื™ืฉ ืœื ื• ืžืกืคืจ ืงื˜ืŸ ืฉืœ ืฉืืœื•ืช,
ื›ืžื• ืฉืื ื—ื ื• ืขื•ืฉื™ื ื‘ื“ื•ื’ืžื” ื”ื–ื•,
06:46
like we do in this example,
159
406201
1317
06:47
we're good to go.
160
407542
1172
ืื ื—ื ื• ืžื•ื›ื ื™ื ืœืฆืืช ืœื“ืจืš.
06:48
Any time OkCupid matches two people,
161
408738
1912
ื›ืœ ืคืขื ืฉOK ืงื•ืคื™ื“ื•ืŸ ืžืชืื™ื ืฉื ื™ ืื ืฉื™ื,
06:50
it goes through the steps we just outlined.
162
410674
2032
ื”ื•ื ืขื•ื‘ืจ ืืช ื”ืฉืœื‘ื™ื ืฉื”ืจืื ื•.
06:52
First it collects data about your answers,
163
412730
2269
ืจืืฉื™ืช ื”ื•ื ืื•ืกืฃ ืžื™ื“ืข ืขืœ ื”ืชืฉื•ื‘ื•ืช ืฉืœื›ื,
ืื– ื”ื•ื ืžืฉื•ื•ื” ืืช ื”ื‘ื—ื™ืจื•ืช ืฉืœื›ื ื•ื”ื”ืขื“ืคื•ืช ืฉืœื›ื
06:55
then it compares your choices and preferences to other people's
164
415023
2985
ืœืื ืฉื™ื ื”ืื—ืจื™ื ื‘ื“ืจื›ื™ื ืžืชืžื˜ื™ื•ืช ืคืฉื•ื˜ื•ืช.
06:58
in simple, mathematical ways.
165
418032
1967
ื”ื™ื›ื•ืœืช ืœืงื—ืช ืชื•ืคืขื” ืžื”ืขื•ืœื ื”ืืžื™ืชื™
07:00
This, the ability to take real-world phenomena
166
420023
2923
07:02
and make them something a microchip can understand,
167
422970
2415
ื•ืœื”ืคื•ืš ืื•ืชื” ืœืžืฉื”ื• ืฉืžื™ืงืจื•ืžืขื‘ื“ ื™ื›ื•ืœ ืœื”ื‘ื™ืŸ,
07:05
is, I think, the most important skill anyone can have these days.
168
425409
3277
ื”ื™ื, ืื ื™ ื—ื•ืฉื‘,
ื”ื™ื›ื•ืœืช ื”ื›ื™ ื—ืฉื•ื‘ื” ืฉื™ื›ื•ืœื” ืœื”ื™ื•ืช ืœืžื™ืฉื”ื• ื”ื™ื•ื.
07:08
Like you use sentences to tell a story to a person,
169
428710
2423
ื›ืžื• ืฉืืชื ืžืฉืชืžืฉื™ื ื‘ืžืฉืคื˜ื™ื ื›ื“ื™ ืœืกืคืจ ืกื™ืคื•ืจ ืœืื“ื,
ืืชื ืžืฉืชืžืฉื™ื ื‘ืืœื’ื•ืจื™ืชืžื™ื ืœืกืคืจ ืกื™ืคื•ืจ ืœืžื—ืฉื‘.
07:11
you use algorithms to tell a story to a computer.
170
431157
2484
ืื ืชืœืžื“ื• ืืช ื”ืฉืคื”,
07:14
If you learn the language, you can go out and tell your stories.
171
434349
3033
ืืชื ื™ื›ื•ืœื™ื ืœืฆืืช ื•ืœืกืคืจ ืืช ื”ืกื™ืคื•ืจื™ื ืฉืœื›ื.
ืื ื™ ืžืงื•ื•ื” ืฉื–ื” ื™ืขื–ื•ืจ ืœื›ื ืœืขืฉื•ืช ืืช ื–ื”.
07:17
I hope this will help you do that.
172
437406
1753
ืขืœ ืืชืจ ื–ื”

ืืชืจ ื–ื” ื™ืฆื™ื’ ื‘ืคื ื™ื›ื ืกืจื˜ื•ื ื™ YouTube ื”ืžื•ืขื™ืœื™ื ืœืœื™ืžื•ื“ ืื ื’ืœื™ืช. ืชื•ื›ืœื• ืœืจืื•ืช ืฉื™ืขื•ืจื™ ืื ื’ืœื™ืช ื”ืžื•ืขื‘ืจื™ื ืขืœ ื™ื“ื™ ืžื•ืจื™ื ืžื”ืฉื•ืจื” ื”ืจืืฉื•ื ื” ืžืจื—ื‘ื™ ื”ืขื•ืœื. ืœื—ืฅ ืคืขืžื™ื™ื ืขืœ ื”ื›ืชื•ื‘ื™ื•ืช ื‘ืื ื’ืœื™ืช ื”ืžื•ืฆื’ื•ืช ื‘ื›ืœ ื“ืฃ ื•ื™ื“ืื• ื›ื“ื™ ืœื”ืคืขื™ืœ ืืช ื”ืกืจื˜ื•ืŸ ืžืฉื. ื”ื›ืชื•ื‘ื™ื•ืช ื’ื•ืœืœื•ืช ื‘ืกื ื›ืจื•ืŸ ืขื ื”ืคืขืœืช ื”ื•ื•ื™ื“ืื•. ืื ื™ืฉ ืœืš ื”ืขืจื•ืช ืื• ื‘ืงืฉื•ืช, ืื ื ืฆื•ืจ ืื™ืชื ื• ืงืฉืจ ื‘ืืžืฆืขื•ืช ื˜ื•ืคืก ื™ืฆื™ืจืช ืงืฉืจ ื–ื”.

https://forms.gle/WvT1wiN1qDtmnspy7