Wed, Aug 19 · 6:00 PM CEST
Welcome to the PyData Berlin August meetup!Make sure to arrive on time! Host :GetYourGuide is excited to welcome you to this month's version of PyData.
Talk 1: From Discovery To Transaction: Personalizing Marketplaces with Intent-Aware Ranking Abstract : In the world of online marketplaces, growth can be a double-edged sword: with a larger supply and user base, making the “perfect match” becomes harder to achieve. For GetYourGuide—a global online marketplace for travel experiences—this meant a single, static experience no longer worked equally well for a diverse global audience. To move past this “one-size-fits-all” limit, we shifted from treating users as one monolithic group to predicting intent in real time.
By analyzing user behavior, we found two common planning modes. Discovery users are exploring what’s possible in a destination. Transaction users return with a clearer goal and are closer to booking.
In this talk, we’ll explain how we built a segmentation framework and a production service that updates these segments in real time using machine learning. We’ll show how these user segments feed into our activity ranking models to personalize results and improve relevance across the search journey. We’ll close with measured impact on engagement and conversion, and what this approach can unlock for other marketplaces moving beyond static user experiences
Speaker : Kaan Isik and Mihail DouhaniarisBio :
Kaan Isik is a Senior Data Scientist at GetYourGuide where he works on the Activity Ranking team. As a member of this team, he focuses on developing and optimizing the learn-to-rank deep learning model that powers real-time activity recommendations for millions of travelers worldwide. His work directly impacts how users discover activities across GetYourGuide's platform, making their travel planning more personalized and efficient.
Mihail Douhaniaris is a Senior Data Scientist at GetYourGuide, where he specializes in marketplace search & ranking algorithms to improve search relevance. His work helps travelers find experiences that match their preferences more effectively. Beyond his role, Mihail is interested in responsible AI, ML observability, and the challenges of deploying machine learning at scale.
Talk 2: Speeding up PyFixest's fixed effects demeaning with new algorithms and Rust Abstract : Fixed effects regressions are ubiquitous in applied data science: they are employed for difference in difference models, switchback designs, variance reduction... Based on the structure of fixed effects, the standard approach to fit fixed effects models at scale can become very inefficient. Therefore, we have developed a new strategy that works extremely well when the default solver struggles. In the talk, I will introduce fixed effects modeling, how it is usually implemented computationally, explain intuitively when/why the default solution can be slow, and introduce our new strategy that is implemented in the rust crate "within" and can be called from Python and R and is already available via the pyfixest regression library.Speaker : Alexander FischerBio : During the day, Alexander works as a data science lead on auction and experimentation topics at trivago. During the night and weekend, he works on open source. Most of his time goes into a package for linear regression - pyfixest - for which, around 4 years ago, we had the stated goal to “make fixed effects regression in Python as nice as in R” before the agents would start writing all of the code. Looking back, we almost succeeded! =)
Lightning talks
***Code of Conduct
