IntRS 2025 presentation at RecSys 2025. Abstract: Recommender Systems (RecSys) have transformed personalized applications by delivering tailored content and
experiences. However, modern Deep Learning RecSys often operate as opaque “black boxes,” offering users no
control over how personalization is shaped. We introduce a novel algorithmic approach to bridge this gap in the
context of visual art recommendation by integrating user agency directly into the RecSys engines. By allowing
users to dynamically adjust facets such as content diversity and popularity, through the use of hyperparameters
implemented as sliders, the system creates a feedback loop where users can actively tune recommendations
while also helping the system to learn about their preferences. This approach ensures that personalization is not
only algorithmically optimized but also user-driven, fostering a balance between automation and human control.
The results of a large-scale user study (n=151) evidenced that sliders enhance engagement and recommendation
quality by promoting meaningful exploration.