A Concept for Consensus-based Ordering of Views

Abstract

High-dimensional data poses a significant challenge for analysis, as patterns typically exist only in subsets of dimensions or records. A common approach to reveal patterns, such as meaningful structures or relationships, is to split the data and then to create a visual representation (views) for each data subset. This introduces the problem of ordering the views effectively because patterns can depend on the presented sequence. Existing methods provide metrics and heuristics to achieve an ordering of views based on their data characteristics. However, an effective ordering of subspace views is expected to rely on task- and data-dependent properties. Hence, heuristic-based ordering methods can be highly objective and not relevant to the task at hand, which is why the user involvement is key to find a meaningful ordering. We introduce a concept for a consensus-based ordering of views that learns to form sequences of subset views fitting the overall users' needs. This concept allows users to decide on the ordering freely and accumulates their preference into a global view that reflects the consensus. We showcase and discuss this concept based on ordering colored tiles from the controversially discussed rainbow color map.