Article by Matfyz Doctoral Student Awarded at International Conference
Patrik Dokoupil, a doctoral student at the Department of Software Engineering at Charles University’s Faculty of Mathematics and Physics, received an award at the international ACM UMAP 2024 conference, held in early July on the Italian island of Sardinia. His paper User Perceptions of Diversity in Recommender Systems earned him the James Chen Best Student Paper Award as the first author. The paper triumphed among more than 90 publications.
ACM UMAP 2024 (32nd ACM Conference on User Modelling, Adaptation and Personalization) is a leading international conference for researchers and professionals working on systems that adapt their appearance or content based on individual user preferences. One of the most common forms of this adaptation is recommender systems, which aim to match users with content that is most interesting to them. Recommender systems are prevalent across various domains, including major multimedia platforms like Spotify, YouTube, and Netflix, news websites, and even small local e-shops.
The best papers were awarded and in the student category, the winning study was authored by CUNI MFF doctoral student Patrik Dokoupil, with co-authors Dr Ladislav Peška (CUNI MFF) and Dr Ludovico Boratto (University of Cagliari).
„In addition to the relevance of recommendations—measuring whether we have correctly matched objects to individual users—diversity (the variety of recommended objects) is perhaps the most important and closely watched metric of recommendation quality. Essentially, we try to avoid situations where all recommended objects are almost the same—such as different parts of the Harry Potter movies. Such recommendations might be relevant for the user, but beyond the first mention, they don't add any new information. Only if the user didn’t know there were additional parts. There are many metrics for measuring diversity, but it is not entirely clear to what extent these metrics reflect how users perceive diversity, and how important small differences in measured values are from the perspective of ordinary people. These are the topics we focused on in the paper,“ explains Dr Patrik Dokoupil.
The findings will be useful in developing and optimizing recommendation algorithms for various target domains, specifically for recommending movies, books, or music.
