We connected with Ahmed Zaidi, a postdoctoral researcher at Cambridge University’s Computer Laboratory, on the networking app Clubhouse. Ahmed spends most of his time thinking about how to solve problems using data, and is particularly interested in applying research in the real world, including dating markets.
Check out his YouTube video “The Algorithms Powering Tinder and Online Dating” here, and then read our interview below!
Tell us about yourself and your interest in online dating
I’m a postdoctoral research associate at Cambridge University where I investigate the role of machine learning in education. I have a PhD and MPhil in machine learning and natural language processing from Cambridge University where I specifically looked at how to model students over time and then use those models to develop an “optimal” teaching strategy. My interest in online dating stems from some of the analogous problems in education and matchmaking. Modeling users and matching them to appropriate educational activities/tasks leverage some of the same techniques that are used to model individuals in search for a partner. As someone interested in human-centric models, the notion of identifying an optimal partner algorithmically is one that really intrigues me.
In fact, outside of my job at the university, I work within the fashion industry where we are yet again trying to solve a similar problem. However, in this particular case the users are shoppers and the life partners are products. Taking a step back, in all three domains, education, matchmaking, and fashion, there is a trade off between time spent searching (explore) and time spent making the most of what you have in front of you (exploit). Striking that right balance is integral to finding the “best” solution.
Can you explain for us the difference between Elo Scores and Collaborating Filtering?
Elo scores is a method of ranking players relative to each other in chess. However, it has also been used in other domains to achieve a similar objective, e.g. ranking country teams in the FIFA World Cup. In the context of dating, it was reportedly used by Tinder as a method of ranking people on their app. The idea behind using Elo scores was that you should be shown individuals who have a similar Elo score to you. There are countless reasons why this is a problematic approach when it comes to dating. One such reason is that the Elo score can be more or less seen as a “desirability score”.
Collaborative filtering is a technique that is used to predict the preferences of users through leveraging the preferences of users similar to them. It makes an assumption that similar users have similar preferences. One of the biggest challenges of collaborative filtering is determining what features are important when trying to determine “similarity”. For example, in the domain of movie recommendations, you might consider including actors/actresses, genre, and directors as possible features to represent users. In the context of dating, you might consider representing users as a function of all the individuals they have swiped right or left on. Therefore individuals who have swiped right on many of the same people are likely to be similar. Using this notion of similarity you can then rank the order in which users are presented with possible prospects.
What about the Stable Marriage Problem and the Gale-Shapley Algorithm ?
The Stable Marriage Problem states that given a finite set of X men and Y women, where each person has ranked their preferences for partners, marry each couple in such a way that there are no two individuals who would prefer each other more than the partners they are currently married to. The Gale-Shapley Algorithm is a method for solving this problem.
The Gale-Shapley algorithm iterates through all the available men according to their partner preferences. If the woman is available then the man and woman become engaged. If the woman is already engaged but the man is ranked higher than the woman’s current partner in her preference list, then the existing engagement is broken and a new couple is formed.
It is worth noting that the Hinge, Bumble and Tinder don’t use these exact algorithms but rather variations of these models. One of the challenges of these dating apps is that the number of men and women is not static and therefore the optimization is sub-optimal.
Why are good photos so important?
Good photos are important because they are by and large the most salient feature of a person’s profile on dating apps. In fact, in each iteration of the app, the images have become larger and there is a minimum requirement for the number of images.
Does the order of people you see in the app matter?
In theory, yes. The order of people presented to you on the app is determined by the matching algorithm, putting the most algorithmically compatible people first.
Connect with Ahmed below:
Twitter: @ahmedzaidi
Instagram: @ahmedzaidi
YouTube
LinkedIn
Website