A while back I wrote IBM Deciphers Personality Types From 200 Tweets. I’ve had countless email discussions with everyone from psychologists to dating site matching algorithm experts and others about the industry’s take on matching systems. While I’m no psychological expert, I have a fair grasp on how the industry thinks and operates.
First, dating sites are simply introduction services. Even the industry moniker “online dating” is false. We don’t date online. We are introduced to people on dating sites.
Millions of people are on dating sites primarily concerned with attracting the lowest-cost customer and retaining them for the longest period of time. Some place more weight on matching systems than others.
The common refrain is that dating sites have low success rates. Yet studies and surveys show again and again that one-in-five people met on a dating site. Is a 20% success rate bad? I don’t think so. Maybe we ought to revise our thinking on what success means.
Dating sites exist to make money. They want you around for as long as possible and they know you’ll probably be back sooner or later. Dating sites live and die by their customer acquisition metrics and keeping members engaged. A few sites are large enough that they can afford to have a team of brainiacs doing research and testing to improve matching efficiency, the rest have pretty basic systems in place.
Finding the Higgs-Boson of matching systems is a pipe dream. What we have is a handful of really brilliant people trying to do the impossible, figuring out human nature, what makes us tick, finding likely candidates and putting them up on our screens.
The time I spent with the Match algorithm folks a few years ago was both eye-opening and mind-boggling. Is their system perfect? Not by a long shot. Are they and a handful of other sites working hard to continuously improve their systems? Absolutely.
Complaining about matching systems is too easy. Look at how stacked the deck is against the dating industry. Millions of profiles to deal with, many without photos. People are deceptive on their profiles about themselves and what they are looking for (not always on purpose). People email back and forth and then do “The Fade”. How many times has a single person received and email and said “Their hair is too short, she is too fat”, or reacting to some deep-seated psychological issue they couldn’t explain if you paid them, like “That guy looks eerily similar to the creepster that grew up next door.”
How the heck are they supposed to match people with cliché-ridden profiles. “On a Friday night I’m at home watching a movie or out with friends”. What does that mean and how can this type of bland information be used for matching?
How about 10 million profiles starting with “I can’t believe I’m trying online dating” as if that has anything to do with anything.
Anyone who knows me knows that I’m fascinated by new and unusual data-points that could be used to improve matchmaking effectiveness.
One of my favorites is “social exhaust”, our Tweets, Facebook updates, Instagram photos and playlists. See also credit card data and far-out stuff like DNA and energy fields and countless other matching data that I’ve come across over the years.
When a reader dissed using social exhaust as a matching fodder, I asked “Figuring out personality type and other factors based on social exhaust is nothing new and there is no reason to think this is valuable data to feed into matching algorithms? How can you possibly say that? Show me some research stating that what we say on Twitter and Facebook has no relative use to online dating.”
They said the only way to revolutionize the Online Dating Industry is using the 16PF5 normative personality test. How does he know this for a fact? What if 16PF5 doesn’t scale well for online dating? 16PF5 could be completely flawed in a large-scale dating application. As mentioned, people knowingly or unknowingly lie on tests, they say they are looking for people they probably shouldn’t be with, and all of the other psychological issues surrounding dating that 16PF5 doesn’t address to my knowledge. So I have three top matches out of 100,000, but who knows if they are the right matches? There is so much bias built into the matching algorithms, lot’s of assumptions based on what exactly?
I’m not arguing against 16PF5 or any other test, I’m critical of the blind “THIS IS THE ONLY THING THAT WORKS” mentality.
I bet if I sat down with the Match and eHarmony algorithm teams today they would show me a lot of impressive data and say, “We think we’re heading in the right direction, but we know it’s an ongoing process and we’re always learning.” Okcupid has said as much in recent news.
The reader went on to say that it’s imperative to calculate personality similarity between users.” To that I say, “Says who?”. What is the definition of personality similarity and who get’s to say only similar types attract? That argument is flawed from the get-go. Millions of couples are married that never would have thought that the person they ended up with was their match. I fell in love with a woman 17 years younger than me, so there’s that. Does 16PF5 or any other modern personality test take this into consideration? I don’t think so.
People will put up with answering questions for maybe 20 minutes. Then you are matched with people who supposedly are a good match. Is there test bias at work? Who defines what a good match is?
Tests tell us about ourselves, which is all fine and good, but what about the thorny issue of what to do with the results?
I assume test creators survey lots of people and figure out “People exhibiting these personality traits tend to pair up with people with these other traits”.
Relevancy is first defined by the algorithm creators. They often start with a premise like opposites attract.
Then by behavioral modeling. Opposites attract vs. similar personality types attracting each other and a million other permutations in between. It boggles my mind thinking about how to categorize what makes a good couple and what datapoints are used to figure this out.
How does the algorithm take into account differences in culture and societies? Ask any site who’s tried to enter the Asian markets after the US for example. The results are not pretty because what works in the US doesn’t necessarily translate around the world. Just ask Match.
What about reducing the number of results returned from an algorithm?
Look at Coffee Meets Bagel. One match a day, but their results are less than satisfying. But they raised a bunch of cash, so it’s got to get better, right? Match of The Day is cute, but the chances of that match being a good one are slim to none. Most people don’t want one match, they want a variety to choose from. Am I crazy for thinking this?
Adult dating sites are much better than casual sites when it comes to matching. It’s also much easier for them to do matching. If I want to have sex, you better look good and be nearby and like… you get the picture. Adult sites get people to fill out profiles with the most intimate details, far more than any dating site.
My firm belief is that people don’t know what they want for the most part and people don’t want to do manual search based on psychological criteria. Never heard anyone ever say they wanted this. What the heck would that search form look like?
Wait its 16PF5 to the rescue! It’s only going to show me optimal results. I call BS on that. No offense to the creators of the test, but it wasn’t created for dating and as humans we are far too diverse, weird and wonderful for a simple canned test to bring us to matchmaking nirvana.
It’s one thing to prioritize results based on psychological/personality factors. It’s another to assume that the test knows what type of people belong with other people.
How come in a decade the 16PF5 people have never emailed me asking for introductions to dating site executives or to make their case? God knows I’ve spent hours on the phone with many other matching system creators.
It would be interesting to hear that eHarmony or some other site was using some sort of 16PF5 variant today. Maybe the top 10 search automated search results are based on 16PF5 already and they add in another couple of hundred results for good measure. Who knows?
What we do know is that dating sites rarely speak about their matching systems. It’s a black box and will stay that way. I wish we had more transparency into these systems, similar to how Google announces their updates.
Zoosk has had $40 Million-dollar quarters and they supposedly have a very strong algorithm. But who really knows? Are they making that much money because of the matching algorithm (successful matches) or because they throw a ton of cash at customer acquisition?
What I hear over and over is that singles don’t care what goes on behind the scenes, they just want to come home from a first date not feeling like a failure. I think singles should be a lot more discerning about the sites they join and how they behave on them, but that’s for another post.
Online dating can be fun, meeting new people is exciting. Given the success of Tinder, why don’t you let humans be humans, browse some profiles, meet some people and maybe go on a date? It’s not like a new test is going to increase effectiveness by much, this stuff is going to take many years to figure out and nobody is even close.
Let’s give it another decade and see what happens.
Until then, happy swiping.