Optimizing Searching and Ranking Systems

This is a past event

94 people went

Every last Friday of the month until September 26, 2019

Location image of event venue

Details

This Friday we'll have two talks followed by drinks.

16:00 Mónica Marrero (Europeana Pro) - Europeana: Challenges in the Search of European Cultural Heritage.

16:30 Harrie Oosterhuis (University of Amsterdam) Optimizing Ranking Systems from User Interactions

========================
16:00 Mónica Marrero (Europeana Pro) - Europeana: Challenges in the Search of European Cultural Heritage.

Europeana aggregates more than fifty million cultural objects from more than 3700 museums, libraries, galleries and audio-visual archives around Europe. We not only offer a digital platform to grant access to our cultural heritage to wider audiences, but we also promote its use in education, research and creative industries.
A good search functionality is key to accomplish these objectives. Here we face the challenge of having a dynamic and heterogeneous collection, to which new object types are added that often require considerable modifications to our search technology. This is the case of our latest collection, almost a million newspaper issues in several languages, from the 17th to the 20th century, and with digitized textual content that can be used to search, as opposed to the search by metadata we offer.
Additionally, in our effort to provide quality data, we have embarked in semantic enrichment with entities (places, time-spans, concepts and agents), and the promotion of multilingual data for both the description and content of the objects. These features contribute to the creation of a richer semantic representation of cultural objects. The new challenge now is how to properly exploit this new information to improve our search capability.
In this talk I will speak about the challenges we are facing in Europeana to achieve these objectives and some of the solutions we have adopted.

========================

16:30 Harrie Oosterhuis (University of Amsterdam) Optimizing Ranking Systems from User Interactions

Ranking models play an essential role for search engines, as their impact on the user experience is enormous it is important that they perform well. Accordingly, ranker optimization: increasing the ranking performance of a system is a vital field of study. In recent years it has become apparent that performing optimization in offline settings, i.e. with human judges/annotators, has its limitations. As an alternative, learning from user interactions has gained a lot of traction in recent years and has become standard practice.
In this talk we will present the latest online learning to rank algorithm: Pairwise Differentiable Gradient Descent (PDGD). PDGD learns by directly interacting with users, and handles interaction noise and biases prevalent in ranking user behaviour (position/selection bias). We will talk about how PDGD can reliably learn from user interactions and the improvements it makes w.r.t. previous online methods.