Love is a physical reaction. When we see somebody, we instantly think “they’re pretty” or “they’re handsome”, and things take off from there. It’s a precious thing, something crucially tied into the continued survival of the human race, and something so personal that it takes most people years to get it just right for them. There’s no doubt about it: love is tied to the very core of what makes humans human. So why is it that today, love is often found through the use of a computer? I’m not talking about falling in love with robots - not yet, anyway. I’m talking about online dating.
There are two big catalysts for the current online dating craze. The first started in the late 1990s, when Neil Clark Warren, a psychologist and relationship advice expert, decided to test a theory he had developed over his 35 years of working as a psychologist and a marriage counsellor. This theory was that a person’s certain characteristics, such as what they like, what they want in a relationship, etc., could be used to predict long-lasting relationships based on their compatibility with others who have similar characteristics. After three years of research, he developed a model of compatibility that would compare the characteristics of whoever used it with their most compatible matches. This would be used as the basis of an online matching system. One that would allow people to meet their ideal mates with a questionnaire and the press of a button. This was how eHarmony was born in 2000.
The other story begins at around the same time. According to OKCupid’s About page, the story starts with SparkNotes. Or, as it was known back then, TheSpark. TheSpark was a website created by four Harvard University students: Chris Coyne, Christian Rudder, Sam Yagan, and Max Krohn. TheSpark had a feature where you could take quizzes and personality tests, and using this feature as a base, the developers created a new experimental site called SparkMatch, which would match you up with people based on the results of these tests. This site got so popular that the founders decided to sell SparkNotes to Barnes & Noble, rebrand this new site as OKCupid and focus entirely on its’ development. Finally, OKCupid was launched in 2001.
After the successes of these brands, companies began to sit up and take notice. More traditional dating sites such as Match.com as well as alternative twists to the online dating formula such as Tinder began to spring up. Today, online dating is a craze. According to The Wall Street Journal, eHarmony received 1 billion dollars in revenue in 2009. IAC, the company that owns OKCupid as well as Tinder and Match.com, had a revenue of 3.1 billion dollars last year. This is a big business, and with that comes a whole diverse market. Sites based on specific religious, racial, sexual and age demographics, sites catering to older couples looking for marriage, sites catering to people who want to treat sex like ordering a pizza, sites that set up where to date along with your potential matches, scam sites that take profiles from legitimate sites without the users’ knowledge and compile them into a package that charges money - it’s a booming and lucrative business, filled with competitors clawing to the top, each with their own algorithms and matching models.
Why is online dating so popular? Simple - it takes something that was so complicated, so mentally and emotionally taxing, that you could read a hundred books about it and still not fully get it, and streamlines it to the point of extreme accessibility. People who are afraid to get out and physically meet others, people whose work or location got in the way of being able to have a normal successful dating life, and people who just didn’t want to put up with trying to guess whether the cute girl sitting across them in science class had a boyfriend or not suddenly have an easy and clear way to meet partners that was really never possible before the internet age. However, despite the popularity of online dating, there’s one thing that most online dating users don’t really take into account that much.
In this episode of inQUERY, we’ll be taking a look under the hood of several dating sites. Using academic studies, third-party information, and the word of the developers themselves, we’ll see how these sites have managed to take the complicated and sticky area of interpersonal relationships and make it into a mathematic algorithm. Note that since there are so many dating sites, we can’t possibly go through them all in the space of a 15-minute podcast, so I’ll be going through the most popular ones and see what they use as models. Afterwards, I’ll talk about how you, the listener, can use these models to your advantage to find the perfect mate for you. If you choose to go that way, of course. Let’s begin.
There are two main models that dating sites use: Survey-based and algorithm-based. First, let’s take a look at OKCupid, which uses a system that most dating sites that require you to do a survey use. According to the developers, OKCupid uses a system where they ask you a series of questions, related to your preferences, your personality and your interests. Whenever you answer a question, you must also check the question as being either a little important, somewhat important, very important or irrelevant, as well as which answers you want the other person to check.
The value of your match percentage is based on these three factors: your answer, how you want others to answer, and how important the answer is to you. Firstly, a numerical value is assigned to each level of importance you assign to a question, as represented in this graph from OKCupid’s website:
When someone answers the way you want them to on a question, they receive points on the same level of importance as the question they answered. This is then compared to the highest possible score you can achieve, and the percentage of that is your match percentage.
For example, say that you and person B are using the site. You are both asked two questions. On the first question, “How organized are you?”, you answer “Very Organized” and you want your perfect mate to answer “Average” or “Very Organized” and mark the question as “Very Important”. On the second question, “Have you ever cheated in a relationship?”, you answer “No, want your perfect mate to answer “No” and mark the question as “A little important”. B answers “Average” on the first question and “Yes” on the second question. Since you marked the first question as very important and the second question as a little important, that gives it a total value of 250 + 1 = 251. B answered the way you wanted him to on the very important question, which gives him 250 points, but didn’t answer correctly on the little important question, which means he doesn’t get that 1 point. Overall, he gets 250/251, which calculates the match percentage to 99.6%. That’s not all, though. There’s also the matter of what B wants his ideal mate to answer and the importance he places on a question. Let’s say that B marked the first question as “A Little Important” and wants his partner to answer “Average”, as well as marked the second question as “Somewhat Important” and wants his partner to answer “No”. He placed 1 importance on the first question and 10 importance on the second one, and of those two you answered correctly on the second one. 10 + 1 = 11, with your answer giving you 10 points, for a total of 10/11 or 91% match percentage. Finally, the site takes you and B’s separate match percentages, multiplies them together and square roots the result for your overall match percentage. In this scenario, your percentage was 99.6 and B’s was 91, so the site would calculate the square root of 91 x 99.6, which is 95%.
For other sites such as eHarmony, the models aren’t quite as cut and dry. The main function of these sites is for users to store their own profiles, and instead of receiving recommendations tailored to their interests, are given profiles to rate. These profiles can be sorted with certain conditions, such as men of a certain age, but aside from that, they are completely random. According to a study conducted by Lukas Brozovsky and Vaclav Petricek in 2007, most dating sites use one of three algorithms. The first is the Random Algorithm, which sorts and predicts profiles with a uniformly distributed random value within the rating scale. The second, the Mean Algorithm, predicts and recommends users with a mean value of all non-zero ratings that a profile has received. For example, if a profile has a mean rating of 2.5 based on the ratings of other users, it will be sorted above users with a lower mean rating and under users with a higher mean rating when searching for mates. User-User Algorithm is one typically used for sites such as Amazon and Netflix that recommend movies based on what the user is watching, but also has its’ uses in online dating. For this one, the user database is searched for users with similar ratings to you, also known as “neighbors”. Ratings of the most similar neighbors for a certain profile are then used to calculate the recommendation for your profile. Another algorithm similar to the User-User Algorithm is the Item-Item Algorithm, only instead of calculating similarities between ratings, it calculates neighbors based on profile similarities. When making predictions for a user, the ratings that the user has received from the most similar neighbors to other profiles are used, and the predictions are sorted based off whose neighbors have rated you the highest.
Unfortunately, User-User/Item-Item aren’t used as often as the first two. User-User has more advantages in that it’s personalized for each user, while in Random and Mean cases, the information inputted by the user is completely ignored. User-User is used in sites such as Match.com, but for a lot of “like/don’t like” sites such as Tinder and Hotornot, Random and Mean algorithms are used. Whether this will be fixed in the future is unknown, but for now, consider sites with Random and Mean algorithms to be of lesser quality than those that use personalized recommendations or User-User algorithms. (https://stephanie-bell-m08b.squarespace.com/blog-season1/e7b7fdc0-dd0b-4ceb-8878-d696b57bb3e8)
At this point, unless you’re a statistics or computer science student, you’re probably thinking “I don’t know anything about math or programming. You’ve done an amazingly succinct and attractive job of summarizing it understandably, but what does any of this have to do with me?” The fact is that, if you want to use a system so that it works fully to your advantage, you’ll have to understand how the system works first. Take this example from Wired magazine: Chris McKinley was a 35 year-old Math Science major at UCLA. He was lanky, had tousled hair and his interests were mostly related to science. These things don’t typically add up to amazing dating material, so he turned to OKCupid as a way to solve his relationship woes. Unfortunately, it wasn’t working out for him - he didn’t have many interests in common with many women in the Los Angeles area, most of his messages were being ignored and the dates he did managed to get weren’t going anywhere. One day, he was up at 3AM using the university’s supercomputer for his PhD dissertation, and when he saw his data compiler in one window and his dating profile in the other window, an idea came to him. Instead of dating like everyone else, he would have date like what he was - a mathematician.
First, McKinley set up 12 fake accounts and created bots to run them. The script used to create the bots would then visit the profiles of his target demographic (in this case, heterosexual and bisexual women between the ages of 25 and 45) and take all the available information from their profiles. Next, McKinley set up the bots to answer questions randomly, then placed the answers of the women whose profiles he had examined and placed them into a database. He got a fellow OKCupid-using friend to agree to install spyware on his computer that would monitor his actions whenever he used the site and make the bots simulate them so that they would appear human to OKCupid’s monitoring system, which banned data harvesting like this. Eventually, he had gotten data from 20,000 women all over the country, and sorted them into clusters using a data sorting system called K-Modes. Using this, he managed to find the two clusters that suited him best - indie-type women in their mid-20’s and slightly older women who held creative jobs.
He set up two profiles, Profile A and Profile B - each appealing to a different cluster. By text-mining the two clusters, he was able to figure out that teaching was a subject of particular interest to both, so he made a profile emphasizing his work as a math professor. For the survey, he answered the questions honestly, but let his computer decide the importance to give each question using an algorithm called adaptive boosting to derive the best weightings and gave the different profiles different importance ratings to each questions based on what his computer told him to do. Finally, he wrote a new bot to visit all the profiles of his top matches. After all this, the numbers began to roll in. His profile was getting 400 views a day, with messages rolling in. However, once he went on actual dates, he found it to be unsatisfactory. By the end of the summer, he had been on 55 dates, with three leading to second dates and one leading to a third. Finally, after 88 dates, he found Christine Tien Wang, a 28 year-old artist attending UCLA. A year later, she was accepted to an art fellowship in Qatar, and during one of their daily Skype calls, he pulled out a diamond ring and she said yes.
Now that was a very adorable story, but what does it mean for you? You’re (probably) not a math genius with the computer abilities of Mark Zuckerberg, so you won’t be able to implement something like this - except you can. At least, that’s what Harris O’Malley of the popular dating site Paging Dr. Nerdlove says. Here’s what you should do: If you’re already a member of OKCupid or any other dating site that requires you to take a survey, clear out all your questions and start from scratch. Don’t answer too many questions - around 200 is the best place to be, any more and you’ll be hitting a point of diminishing returns. Answer each question honestly, though that should be obvious. Never answer below “Somewhat important” when weighing the importance of a question, and if a question doesn’t matter, just skip it. Finally, pick the right questions. This one is a bit trickier. You’ll have to go through your matches, click “The Two of Us” tab, which tells you which questions both of you have answered, and look at the “Questions She Cares About” in the sorting bar. Answer several of these questions from your most compatible matches, and you’ll eventually get a handle of what people you’re interested in care about the most. (https://stephanie-bell-m08b.squarespace.com/blog-season1/4a75d97e-f338-4281-8dc4-d31c1c9e6ccb)
These steps may seem a bit unnecessary at first, but they’re vitally important if you want to find your match more efficiently and not waste time on people who really aren’t your thing. The higher your match percentage is, the better off you’ll be.
Well, that’s all the time we have today. I hope you’ve managed to learn a thing or two about the changing nature of love and how to take advantage of it to find the best person for you. If you’re one of those lucky people who already have a boyfriend or girlfriend, well, first of all, screw you, and second of all, I hope you found something interesting anyway. Now all that’s left is for scientists to collect the data of our potential matches, store them in sentient robots and sell them to whoever wants them. Maybe next year. See you.
Brozovsky, L., & Petricek, V. (2007). Recommender system for online dating service. arXiv preprint cs/0703042.
O’Malley, H. (2014, January). How to Hack OKCupid. Paging Dr. Nerdlove. Retrieved from http://www.doctornerdlove.com/2014/01/how-to-hack-okcupid/
Poulsen, K. (2014, January). How a Math Genius Hacked OKCupid to Find True Love. Wired Magazine. Retrieved from http://www.wired.com/2014/01/how-to-hack-okcupid/all/
OKCupid. (2015). OKCupid Support. Retrieved from http://www.okcupid.com/help
Geron, T. (2010, July). The $100M Revenue Club: EHarmony Captures Hearts Of VCs. The Wall Street Journal. Retrieved from http://blogs.wsj.com/venturecapital/2010/07/12/the-100m-revenue-club-eharmony-captures-hearts-of-vcs/