Online dating pioneer eHarmony is betting on machine learning to find long-lasting relationships for swipe-fatigued singles.

The fee-based service believes the core premise that the company was built on 17 years ago — to help you find a longterm partner — is still what differentiates it from its app-based competitors that are quicker to set up and free to use. 

I think what differentiates us from the rest is we are really trying to add value to the whole online dating proposition,” says Prateek Jain, VP technology at Eharmony.

Being an early player in the online dating space has given eHarmony the capital and data required to improve its technology and matching capability, Jain said. 

“We are an incumbent but that also gives us the early commercial advantage that we have been able to invest in our technology a lot more and make it more sophisticated.”  

“A lot of work is being done to present the core premise in a very contemporary and modern fashion.” 

Prateek Jain, VP technology at Eharmony.
Prateek Jain, VP technology at eharmony.

As well as improving its mobile apps and user interface, eHarmony is using machine learning and data science to figure out its users’ preferences and hopefully make more successful matches.

Jain says “there’s a fatigue that builds up” with apps that involve endless swiping and little communication. In fact, eHarmony CEO Grant Langston is hoping all those millennials looking for long term relationships on Tinder will eventually get tired of swiping and head over the eHarmony.

“All a lot of these new dating sites are doing is [filtering by] distance, location, age… We are trying to make matches on a much deeper level,” Jain said.

Given that eHarmony doesn’t ask its users if they end up going on a date, the machine learning models are optimised for two-way communication. That means the ideal outcome — according to the machine — is someone sending a message to their match and getting a reply. That’s the best indicator that both parties are happy with the match, Jain says.

He explained, it’s like picking a movie Netflix has recommended you watch, but the movie has to like you back.

How to make (and optimise) a match

Eharmony uses two techniques to match singles. The initial match is based on compatibility. This measure is determined by the extensive questionnaire users complete when they join the site as well as patented mathematical models.

This step is designed to match people who are similar and will be (again, hopefully) compatible for the long term.

But, as Jain explains, “I could find you the most compatible person on the planet but what if they are not attracted to you?”

Eharmony then uses machine learning, which it calls affinity matching, to learn about behaviour on the site as an indicator of what you like. For example if you are more likely to communicate with a match that has more than 500 words on their profile the next batch of matches will include more complete written profiles. Or, in terms of physical appearances, analysis of photos lets the ML know if a users likes blondes or beards. 

Testing the model

Eharmony currently runs 20 different affinity models. But how do they know if what they’ve built actually works?

Jain explained his team will run an A/B test using the model that they have trained and a model that just predicts random outcomes and compare the results.

“If the random model is producing almost similar results as your production model, then really your production model is not doing much,” he said.

“That’s how you keep the data sanity and ensure your models are on the right path.”

Previous post

Reaching Gen Z and Managing Hyper Growth With Unidays CEO Josh Rathour

Next post

Twilio opens Australian office to drive growth

Join the digital transformation discussion and sign up for the Which-50 Irregular Insights newsletter.