De Econometrist neemt een statistische kijk op de wereld.
Although it seems as if mobile applications for online dating are mostly about connecting new people, the mathematics used behind the scenes is intriguing. What do we know about the algorithms used for these apps and what does the app know about us? And, more importantly, how is our online dating life influenced by this information?
With the availability of online dating applications, it is getting more and more easy to meet and date new people. For example, using Tinder, you can see the profiles of people around you. Based on their pictures and biography, you can choose to either swipe them right or left. When swiping someone right, you ‘like’ him or her, and once this person swipes you right as well, a match is established and you can start chatting. It seems as if all people in and around your neighbourhood show up in your feed, in a somewhat random order. However, this is not at all as random as one might think. Behind the quite simple concept of the app Tinder, there is a much more sophisticated algorithm determining which people will and will not be shown as a potential match for you.
The details behind these algorithms are, unfortunately, kept secret. Luckily, some manners are published about the way Tinder determines your potential matches. For example, we know that a so called ELO-score is used. Although this seems to be a plain measure of attractiveness, Tinder’s co-founder Sean Rad claims that it is more than just that. “It is not just how many people swipe right on you, it is very complicated. It took us two and a half months just to build the algorithm because a lot of factors go into it”, he says in an interview with Fast Company. The ELO-score depends on the kind of profiles you like, or swipe right, and on the kind of profiles you like. For instance, if someone with a much higher ELO-score likes you, this has a positive effect on your personal score. Therefore, it can be said that roughly, you will be classified in a cluster with profiles with an ELO-score comparable to yours. This has as a result that you will mostly see profiles of people that are approximately as ‘attractive’ or desirable as you are, and that your profile will mostly be seen by these people.
Another trick that Tinder uses to adjust your potential matches to your personal preferences, is the relatively new update Smart Photo, of which each user can choose to turn it on or off. When turned off, a user chooses which of his or her photos will be firstly shown to other users. However, when turned on, Smart Photo determines which of your pictures leads to the most matches when shown first. Besides that, it might change your top photo according to the preferences of other users. For example, we consider the case of a female Tinder user who swipes right relatively many profiles of men with a picture of their shirtless selves, proudly carrying a fish they just caught. If a user has four pictures, of which one satisfies the above stated description, Tinder makes sure that this picture is shown first to the female Tinder user, in order to increase the chance of a match. However, if the female Tinder user has a friend that never likes these kinds of fishers, the Smart Photo option makes sure not to show such a picture as someone’s top photo. If this someone will even be shown as a potential match to this friend at all.
With the launch of Smart Photo, Tinder saw an increase up to 12 percent in matches for profiles who turned this feature on. Considering the fact that Tinder has around 9.6 million daily active users throughout the world, that together account for 1.4 billion swipes every day, this leads to many dates that would never have occurred without the availability of online dating apps. Besides improving the connection between new people however, which is the main aim of Tinder, a feature like Smart Photo also shows how Tinder is not only a social network, but a big data company as well. With Smart Photo, it is clear that Tinder collects a lot of data on its users and their personal preferences. Although it is not the case already, one can imagine that this information can for example be used to adjust advertisements to the personal preference of its users.
Where Tinder mostly uses pictures and a short biography, dating application OkCupid, which has 7.3 million messages sent every day, needs more information to match profiles to each other. In your profile, you can answer some out of thousands of random questions, regarding for example your religion and your willingness to discuss politics during a date. Given your answers and your opinion on how important someone’s answer on each of these questions is, OkCupid shows you potential matches, who you can like, dislike or immediately start chatting with. For each of the profiles shown to you, OkCupid gives you a percentage on how good your profiles match, based on your and their answers to the questions. Since you only answer a part of all the questions in the app, chances are that one of your potential matches has only given answers to some completely different questions. Therefore, the match-percentage given by OKCupid might differ largely from the, hypothetical, case where you both answered all available questions. Therefore, this match-percentage can be seen as quite questionable.
A question that rises with the information on how potential matches are determined by OkCupid, is whether, and how, you can influence this. Someone who did some extreme research on this is Chris McKinlay, an American mathematician. At the age of 35, after having been on OkCupid for a while, he was not satisfied with the match-percentage he had with women in his feed, and with the amount of time and the work it took him to find someone appropriate to go on a date with.
As a solution for this problem, he hacked the app and collected data on how women answered all
questions given by the app. With this information, he divided all women in seven different clusters. Then, he decided which of these clusters contained the kinds of women that satisfied his preferences. Based on what questions most of these women found important, he adapted his profile. As a result, McKinlay suddenly founds himself as a >90% match for over ten thousand profiles! Women started approaching him over the app more than ever before and he started dating frequently. Over the summer he went on more than 55 first dates, of which only three led to a second and only one to a third date. At first date number 88, he met Christine Tien Wang, to whom he is engaged now. To read the whole hacking process of McKinlay, see this link.
The way McKinlay approached OkCupid to find his perfect match might be seen as too complex for most of us, but are there other ways to optimize the probability of finding the match you are looking for? According to Jessica Carbino, sociologist at Tinder, there are some simple steps to increase your amount of matches. First, smiling on your profile picture can already increase your number of likes by 14%, where a smile towards the camera even causes a 20% increase. Wearing a hat however, decreases your probability of being swiped to the right by 12%, and wearing glasses by 15%. And what about how to act after a match has been made? Dr. Carbino says that sending a GIF to your match might be thebest move to make: the probability to get an answer to a GIF is 30% higher than to a simple text message. An exception of this is the ‘how you doin’-GIF of Joey, a character from the comedy Friends, which is seen as the cliché Tinder pickup line.
Besides of helping you improving your profile, the data collected in these researches can be useful for many other purposes. For example, we can find out how age preferences differ between men and women over time. According to Christian Rudder, founder of OkCupid, women until the age of 40 prefer men of their own age, whereas men of age 25, 30 and 50 all prefer women aged 20. Another interesting aspect is the difference in preferences among races over time. The number of OkCupid users who answered ‘Yes’ to the question whether they strongly preferred to date someone their own race decreased from around 40% to 30% between 2009 and 2014. However, statistics on the matches between OkCupid tell us something different.
In the table below, one can see how different races are preferred compared to the average. For example, the top-left corner in the 2009 table tells us that Asian men see an Asian woman as 11% more attractive than the average woman. Comparing the two tables, it can be found that women, more than men, prefer to date someone their own race. Furthermore, this preference has increased over the last years, which is in contradiction to what the users answer when they are asked right away. Especially black female users and both Asian and black male users are seen as least preferred by users of another race.
It may be seen as shocking that apps like Tinder and OkCupid collect and use all this information, but is it really alarming? All information collected is given up by the users itself and is in fact open to use by the apps. Furthermore, as Christian Rudder says, OkCupid is suggesting people in the same way that Netflix gives suggestions on what movies you may want to watch, based on the movies you have been watching before. By collecting and observing data, these suggestions can be more and more adjusted to your personal preferences, increasing the probability that you find your perfect match. Until that time, just keep smiling in the camera and refrain from sending Joey-GIFs.
Dit artikel is geschreven door Marleen Schumacher