A Design Space and Exploration Tool for Visual Interfaces in Recommender Systems

Recommender systems are a ubiquitous part of the online experience, yet their interfaces often appear as black boxes, making it difficult for users to understand, explore, and influence recommendations. Prior work has shown that interactive visualizations can improve transparency, trust, and user experience, but existing solutions are highly diverse and lack a unifying structure. In this paper, we present a design space for visual interfaces in recommender systems that captures recurring interface properties across WHY, WHAT, and HOW dimensions. Derived through a multi-stage empirical process, it was applied to deconstruct 58 visual recommender interfaces into 96 distinct views. Based on this design space, we further introduce an interactive exploration and analysis tool that enables users to search, filter, and compare interface solutions and to analyze recurring combinations of building blocks across challenges, data characteristics, and visualization and interaction techniques.