Dating site OkCupid success cheats: "Experiment" with users

Intermediary transaction SEO diagnosis Taobao guest Cloud host technology Hall

OkCupid is a very popular dating site abroad, and they also have their own app apps. This digest is from its website blog, the author of the company co-founder Christian Rudder.

I am the first person to say that: we will become very popular, we will be a lot of people to lead the red line, we will achieve a lot of things. But OkCupid does not seem to really know what it is doing, of course, other sites may not know. What we do, not many people have done before, so the industry experience is not much, and there is no blueprint and so on for reference. In fact, most of our ideas are very bad, and even if some of the ideas, can actually do better. So how to solve all these problems? The answer is to do the experiment. The experiment lets you know anything, such as the following picture, the young child wants to experiment to see if the potatoes will cry.

But lately we've noticed that people don't like to be tested, for example, when Facebook had "experimented" with users when they launched their news feeds, but it turned out to be objectionable, and it even involved the Federal Trade Commission (FTC). But guess what, everybody, PS. If you use the Internet, everyone is actually the object of the experiment, even at some times the same person may be hundreds of experiments, almost every site has done so, otherwise where so many successful sites?

Here, I'd like to introduce some of the very interesting experiments that OkCupid did before.

Experiment one: Is love blind or should there be a goal?

OkCupid Company has been created for more than 10 years, the epitome of our company can be described in an old saying, that is two steps, step back. We had a very creative idea to create an app that we could use for a blind date, so we spent 1.5 of our time concentrating on the app, but it backfired, and it was just six months behind the Apple App Store.

Of course, we are all "geniuses" in the OkCupid company. On the day of the release, we decided to celebrate, on January 15, 2013, we made it "love is Blind Day," and we decided very boldly to do an experiment that removed all the photos on the OkCupid.

During the celebration, all the metrics on our site fell, as shown in the following figure:

But if we compare "Love is Blind Day" with an ordinary Tuesday, we find some interesting things in our website and applications without photos in seven hours:

1. The rate at which people respond to short messages for the first time is 44% higher than usual.

2, the dialogue between people's content has become deeper.

3. The frequency with which people exchange contact information (including cell phone numbers and emails) becomes faster.

4. In short, OkCupid works better.

When we were doing the experiment four o'clock in the afternoon, all the photos were restored, at that time, there were 2200 people in the process of "blind" conversations with each other, but when everyone saw each other's photos, the conversations slowly disappeared. Good things have slipped away, of course, the bad things are Gone with the wind. This feeling, like in the Midnight Bar, suddenly all the lights are open. The image below is a long image of the conversation when we restore the user's picture, and from left to right it turns out that the date of the conversation is getting shorter.

The experiment made me very curious, so I went back and looked at some of the user data, which were actually used by our dating app, and soon, I found a simple thing that once they really loved each other and started dating, they would be very happy and their partner would be good-looking. , seems to have become no longer important. The image below is a comparison of women's looks (in fact, men and women have very similar views on this issue):

Strangely enough, I found one more thing. The more good-looking women seem unhappy on dates. And from my actual experience, those who like to pick up the Congole men are often "bastards." Perhaps this explanation is not clear enough, let's say in other words. Good-looking women tend to be "cold", and if someone sends them a message, they generally don't respond. And good-looking men tend to be more enthusiastic, and if someone sends them a message, they will respond positively. In fact, the user's online response also confirms this view, as shown in the following figure,

Basically, people are really "superficial" because they are generally influenced by technology.

Experiment two: So, what is the value of a picture?

All dating sites have a function of rating user data. OkCupid the original system provides users with two different evaluation factors, allowing users to evaluate each other, two factors one is "character", the other is "appearance."

I found the past system and made a screenshot (as shown below), for the first four years of our site, the "Loading" icon has always been covered by the user's front photo, anyway, our voting system is in the image below, and I marked it with a red arrow:

In our opinion, a person's looks may not be very classic, or handsome to disturb the party Central Committee, although he/she looks very ordinary, but can be very cool people, we hope we can realize this, and from the beginning of OkCupid, we must realize this. Perhaps the only thing that makes us feel a little weaker than the bad HTML pages we write on our site is the understanding of human nature.

In our site's backup database, I dug up some data. In the image below, each point represents a person. The horizontal axis represents the "appearance" of the score, while the longitudinal axes represent the "character" rating.

In short, according to our users, "appearance" and "character" are actually the same thing for them. Don't believe it? Well, look at this picture, the young and beautiful female user, her "character" reached 99 points.

Users of that sort, by the way, they/their data do not need to contain words at all, of course, these people are very cool, but also can chat with them, and date.

Later, we put the previous two "indicators" to be removed, replaced by an indicator, directly to the user to see photos. Of course, this time we used some users to do the "experiment", we give some users only look at the photos, and to the other part of the user with a text presentation. Results each user's data produced two different scores, one of which was produced by "text data plus photos", and the other was "just looking at photos". The diagram below is the contrast diagram. Of course, each point also represents a user.

The result is a glance, as the old saying goes, a picture is better than words. In other words, if you write a bunch of personal introductions, you might as well match up a picture.

Experiment III: The strength of the proposal

Now, the last question that OkCupid has left is, is it a good thing to date online? Through all of our internal evaluations, the "matching proportions" we calculate for our users are well on the way to predicting appointments. These calculations involve a lot of things, including the number of successful messages sent, the length of conversations, the actual exchange of contacts, and so on. But behind our creativity, there seems to be always a possibility that our site works, perhaps simply because we tell the user that it works, and maybe people like each other just because they think they should. This may also have some reason, otherwise there will be so many people still buy Jay album.

But in order to test, we did "experiment". We found some dating results that weren't very good, about 30% of all matches, and then we told them they were really, really good at each other. Not surprisingly, when people are told they are able to live in harmony with each other, they are more willing to give each other the initiative to send messages and communicate with others. But in fact, this is a website that teaches them to do so through technology.

However, we will take a step further in the analysis. We ask: is the matching ratio shown only to motivate people to send messages, and if we offer more advice, will people really like each other? From the results of our current experimental tests, yes, really!

When we tell people that their match is a very good pairing, they react as if they really found something they liked, and even sometimes they feel a little uncomfortable with each other, but still feel that the message is true.

In our internal "real exchange" assessment, there is an indicator that we believe that the two sides must have four short messages in order to be a real exchange. Although the data is a bit messy, it also illustrates the fact that the more couples interact with each other, the greater the chances of a successful date. Similarly, if both parties exchange contact information, the likelihood of success will increase greatly.

Such a result, also let us feel a trace of worry. Maybe our pairing algorithm sucks, but it gives the user some advice that they can be together. So we did an "experiment" from another angle: We told the users who were very good paired before, that you weren't really the right person to see what would happen.

The following picture shows the results of our experiment:

From the above figure, as you can see, the horizontal axis is the match index that two people are told, the number axis is actually the matching index of two people, and the color part is the actual communication situation between the two people. The ideal state is the "lower right corner" state, that is, when the two sides have a good pairing, but also be told that each other is their "true love", the probability of success is the greatest.

After so much analysis, you may feel that the chances of finding another half on the OkCupid are still great.