Recommending Paintings in Web Art Gallery with Adjustable Popularity and Diversity

The cold start problem remains a major challenge in visual art recommendation, where limited user feedback often forces systems to rely on content-based filtering. While effective with sufficient data, content similarity-based recommendation can reinforce filter bubbles, narrowing user exposure to mainstream content. Popularity and diversity are both critical factors in recommendation systems, as they impact the visibility of niche items and overall user satisfaction. Yet, existing platforms often rely on popularity-centric algorithms that may discourage exploration and overshadow lesser-known items. To address this gap, our work investigates whether users' preferences for popular and diverse recommendations remain stable over short sessions of recommendation. We propose an interactive, user-Adjustable mechanism allowing individuals to control the balance between mainstream and novel suggestions in real-Time. We implement this approach within a Web gallery recommendations. Through user study, we examine changes in user behavior. Our findings suggest that while many users initially gravitate toward popular and diverse content, providing controls encourages later adjustments and exploratory behavior. This highlights the need for cultural institutions to move from a tightly managed centralized model to offering users greater affordances for managing the popularity and diversity of personalized recommendations.