Cluster Sculptor, an interactive visual clustering system

Abstract

International audienceThis paper describes Cluster Sculptor, a novel interactive clustering system that allows a user to iteratively update the cluster labels of a data set, and an as-sociated low-dimensional projection. The system is fed by clustering results computed in a high-dimensional space, and uses a 2D projection, both as sup-port for overlaying the cluster labels, and engaging user interaction. By easily interacting with elements directly in the visualization, the user can inject his or her domain knowledge progressively, crafting an updated 2D projection and the associated clustering structure that combine his or her preferences and the manifolds underlying the data. Via interactive controls, the distribution of the data in the 2D space can be used to amend the cluster labels, or reciprocally, the 2D projection can be updated so as to emphasize the current clusters. The 2D projection updates follow a smooth physical metaphor, that gives insight of the process to the user. Updates can be interrupted any time, for further data inspection, or modifying the input preferences. The interest of the system is demonstrated by detailed experimental scenarios on three real data sets

Interactive Clustering

“…Finally, interacting with clustering results as described in several of the papers consist of different combinations of adapting various parameters in more or less independent ways (refer to Section 3.2). Bruneau et al [23] uses the distribution of the data to transform the high-dimensional clustering, updates the two-dimensional (2D) embedding by dissimilarity transform, and allows users to limit operations to a subset of data. In Xu et al [113], users interact with node-link diagrams, adjacency matrices, and tree-maps to refine clusters, either by directly interacting with the visual representation (e.g., remove a node from a cluster or relocate a node to an appropriate cluster) or by updating the similarity measure.…”

“…In Kwon et al [65], the user's expertise guides the clustering algorithm toward the right size and type of clusters. Bruneau et al [23] allows the user to "easily inject his or her domain knowledge progressively." The authors in Huang et al [59] use a visual representation of the clustering tree for refinement and validation purposes, because, based on their knowledge, the users accept or reject partitions.…”

“…For instance, in Basu et al [15], Chen and Liu [28,29], desJardins et al [35], Guo et al [49], Turkay et al [100], the user can perform drag-and-drop operations to edit clusters. But other interaction types are also available, such as through changing the parameters used in a contingency table or matrix (e.g., Balcan and Blum [13], Bruneau et al [23], Hossain et al [54], Looney [70]). Some tools, however, only support merging (e.g., Looney [70]), where the user is first presented with a large number of clusters that can either be removed or merged.…”

Clustrophile 2: Guided Visual Clustering Analysis

“…Matchmaker [20] builds on techniques from [34] with the ability to modify clusterings by grouping data dimensions. ClusterSculptor [27] and Cluster Sculptor [4] enable users to supervise clustering processes. Schreck et al [33] propose using user feedback to bootstrap the similarity evaluation in data space (trajectories, in this case) and then apply the clustering algorithm.…”

Rate-Distortion Theory for Clustering in the Perceptual Space

“…Matchmaker [22] is a visualization technique that makes it possible to split and individually combine a multidimensional dataset into several groups of dimensions, run clustering algorithms on these groups separately and then visually compare the results. ClusterSculptor [23] is a tool that enables users to supervise clustering processes in various clustering methods. Schreck et al [24] proposed a framework that enables the user to visually monitor the clustering process and control the unsupervised self-organizing map algorithm at an arbitrary level of detail.…”