Integrating user visual preferences into adaptive dashboards: a vignette study of encoding choices.

Details

Supervisors

Vanderdonckt, Jean

Faculty

Louvain School of Management

Degree label

Master [120] : ingénieur de gestion, à finalité spécialisée

Abstract

Dashboards have become a central tool for data-driven decision-making, yet they are usually designed once and delivered unchanged to every user. Adaptive dashboards address this by adjusting their content and presentation automatically to each user. Such systems decide how to display data from user profiles or observed behaviour but very rarely take users' visual preferences into account, despite their known impact on system adoption. This thesis explores how users' visual preferences can be effectively integrated into an adaptive dashboard such as the one developed within the SYMBIOTIK project, by answering how visual preferences are distributed across users and how they should be weighed against the accuracy with which a presentation is read. A vignette study was conducted on the Crime Investigation Dashboard of the SYMBIOTIK project. Seventeen participants expressed 1,024 paired preferences between variants of the same visualisation, differing in the visual choices the system itself adapts. Preferences were estimated using Bradley–Terry models, then crossed with Mackinlay's rankings of perceptual effectiveness to produce preference × effectiveness matrices. Preferences prove to be shared on broad choices (whether to adapt the presentation at all, and how much information to show). However, results do not provide enough evidence at the finer level of individual visual choices, where each participant made too few comparisons for genuine agreement to be told apart from randomness. Although they are based on important assumptions, the matrices show where preference and effectiveness point in the same direction and where a system should choose between the two. Among the visual choices tested, colour is the one users consistently prefer, whereas changes to the size of elements or to the background are not. Since participants judged appearance without ever using the system, these results sketch a decision space rather than establishing effects, and specify what a confirmatory study would need to measure.