Category: YesnomayB

OPW INTERVIEW — Apr 11 — The founder of yesnomayB and Netflix Prize finalist, Gavin Potter, have joined forces to bring behavioural matchmaking and personalization to online dating and social networking sites. Here’s our interview with the Founders of Intro Analytics. – Mark Brooks

What is the founding story of IntroAnalytics?
Gavin: For the last couple of years, I’ve been working in collaborative filtering loosely based around the $1m datamining competition called Netflix Prize for the best collaborative filtering algorithm that predicts user ratings for films, based on previous ratings. Nick got in touch with me and said, “do you think we can apply the same type of approach to internet dating?” So we gave it a go. We found that we could get some very accurate results just by looking at how people browse a particular dating site.

So the idea of this is to help people get better results from their browsing?
Gavin: That’s absolutely right. Most dating sites run and work on profiling information. If you can use profiling information and browsing information you can get much more accurate recommendations.

If I’m only clicking on images of women with red hair, as a result you’re going to give me profiles of women who only have red hair?
Gavin: That’s almost right. What we try and do is capture a person’s actual behavior rather than base it on the information in the profile. So, if despite what they say on the profile, they only click on women with red hair than our algorithms will tend to recommend women with red hair. If they only click on women with an income above a certain level, then our algorithms will tend to recommend women with an income above a certain level etc. This is the beauty of this technique, it captures what people actually do rather than what they say that they want. When you combine the two together you get a better result.

Nick how did you get involved with IntroAnalytics?
Nick: We launched a dating site called yesnomayB.com about 2 years ago. The whole premise of the site was to make a first-impression choice of people based on a high-quality photo with keywords. We then use a collaborative filter algorithm to recommend new dates to you (just like Amazon recommends books on what choices you have made in the past). We found this an extremely challenging task and I was starting to lose hope until I met Gavin in March last year. He presented me with incredibly accurate results based on user browser behaviour.

Further, it was more amazing to see very good 2-way recommendations being presented to members who had never clicked on each other before.

Does it also look at the text of the profile?
Gavin: It only captures what people are looking at – if they are basing their results on profile information it will capture that. If they are basing their browsing behavior on attractiveness it will capture that. It is true, however, and slightly to our surprise that after the location of the potential date, attractiveness appears to be the most important characteristic regardless of the information provided. We’ve now worked the algorithms on dating sites that contain substantial profile information as well as websites that contain very little and this finding doesn’t seem to change.

How many clicks will it take until you really start identifying what this person likes?
Gavin: You can start from just one click but clearly that’s not going to give you a very accurate result. As soon as we identify 10 individuals that someone has expressed an interest in, we can start getting some quite accurate results in terms of predicting who they would be most interested in.

In terms of identifying their interest, is it both just clicking on images or also initiating communication?
Gavin: You can use anything that expresses an interest. Right now we are working with a couple of sites who are looking at combining browsing behavior with messaging or going on a date with someone.

You said it’s a mathematical algorithm. How do you quantify its results?
Gavin: There are 3 ways. The first way that we’re trying to quantify is to see if the users are interested in using the recommendations that come out of the algorithm. The second way is when we are looking at whether people have messaged. We’re trying to predict whether someone will message someone and then we see how accurate those predictions are. The third way, we are still experimenting with, is to look how many people we’ve converted based on using this technology.

If some new person joins a site, we can quickly identify who that person might be most interested in and we can then make some accurate recommendations to those people.

Are there any other sites using your service at this point except of yesnomayB?
Gavin: There are 3-4 companies evaluating our technology at the moment.

The technology right now is at yesnomayB.com. Nick, what are your observations so far?
Nick: We can say with certainty that our users are definitely using the recommendations and we have received som every positive feedback. We believe that recommendations will play a major role in converting our members to subscribers.

If you look at other industries like Amazon and Netflix: Amazon’s recommendations create 30-35% of its overall sales. I also know that Netflix rents two thirds of it’s movies through recommendations.

Recommendation and discovery of products and services is complex but easy to implement with the right algorithm – all dating sites of the future should be doing it as users and dating sites both win.