The 15th ECML PKDD Discovery Challenge: Recommending Given Names

20DC13 Logo

You probably already know recommender systems from online shops: "Customers who liked this product, also showed interest in...". Within the challenge, new approaches for name recommendation systems are developed and applied ("People who like 'Peter' also like the name 'Paul'") and integrated in Nameling.

All these recommendation requests are anonymously sent to all participants of the challenge. Their recommender systems have to provide a list of recommended names in real time. The system with the most positive feedback will win the challenge.


Here are our contestants - the participants of the name recommendation challenge in the quest for your perfect name!

Ben Letham

Ben Letham

I am a PhD student at MIT studying operations research, which is an area of applied math that focuses on using data and mathematical models to make better decisions. In my research, I work on applications in medicine, internet search, and sales.

My Nameling recommender system is based on two main ideas: user similarity, and name combination patterns. When you ask for a name recommendation, the recommender system looks at the list of names that you have entered or clicked on (your history), and tries to find other users with a similar history. The idea here is that users that are similar to you will probably be useful for determining names that will be of interest to you. On top of finding similar users, we look for patterns of names. For example, it may be that users who liked the name Sheldon were also very likely to like the name Leonard. We would notice this pattern, and then if you entered Sheldon, would recommend Leonard. There are many possible patterns we could use to make recommendations, and we focus on patterns that are strong in the users that are most similar to you. Every name on your recommendation list comes from the histories of users that have entered similar names.

You have probably already decided to name your child Ben, but if you still aren't sure, then I hope my recommendations will be helpful to you!

Katja Niemann

Ben Letham

I'm a researcher at the Fraunhofer Institute for Applied Information Technology in Sankt Augustin (Germany) and a PhD student at the RWTH Aachen University. My research is all about making web users happy. In order to do so, I analyze the usage of objects (e.g. text documents, movies, or names) in web portals and use the found relations to improve the recommendations the users receive.

My Nameling recommender system is based on the assumption that users want to receive recommendations for names that are similar to those they already looked for or even stored as favorites. The problem is: How do we decide if two names are similar or not and what is meant with similarity? We define two names to be similar if most of the users who like one of the names also like the other one. Thus, an obvious approach would be to calculate which names are often searched for or stored as favorites by the same users. However, with this approach we can only find a few similarities. Though, my Nameling recommender system analyses not only which names are often used together but also which names (that maybe were never searched for by the same user) share similar names. For example, a lot of users search for the names “Anna”, “Emma”, and “Laura”, thus, these names are likely to be similar according to our definition. The name “Lia” was searched for only a few times but often together with “Anna” and “Emma”, thus, “Lia” seems to be similar to these two names. Additionally, we can assume “Laura” und “Lia” to be similar, because they are both similar to “Anna” and “Emma”, even if not one user searched for both names.

Have fun finding a name for your baby!

Our Sponsors

We greatly appreciate the help of our sponsors!