Clustrophile 2: Guided Visual Clustering Analysis

Abstract

Data clustering is a common unsupervised learning method frequently used in exploratory data analysis. However, identifying relevant structures in unlabeled, high-dimensional data is nontrivial, requiring iterative experimentation with clustering parameters as well as data features and instances. The number of possible clusterings for a typical dataset is vast, and navigating in this vast space is also challenging. The absence of ground-truth labels makes it impossible to define an optimal solution, thus requiring user judgment to establish what can be considered a satisfiable clustering result. Data scientists need adequate interactive tools to effectively explore and navigate the large clustering space so as to improve the effectiveness of exploratory clustering analysis. We introduce Clustrophile 2, a new interactive tool for guided clustering analysis. Clustrophile 2 guides users in clustering-based exploratory analysis, adapts user feedback to improve user guidance, facilitates the interpretation of clusters, and helps quickly reason about differences between clusterings. To this end, Clustrophile 2 contributes a novel feature, the Clustering Tour, to help users choose clustering parameters and assess the quality of different clustering results in relation to current analysis goals and user expectations. We evaluate Clustrophile 2 through a user study with 12 data scientists, who used our tool to explore and interpret sub-cohorts in a dataset of Parkinson's disease patients. Results suggest that Clustrophile 2 improves the speed and effectiveness of exploratory clustering analysis for both experts and non-experts.

t-viSNE: Interactive Assessment and Interpretation of t-SNE Projections

“…In contrast to t-viSNE, it does not support further exploratory visual analysis tasks after the layout is selected, such as optimizing the hyper-parameters for specific user selections. Clustrophile 2 [23] contains a Clustering Tour feature to partially assist users in immediately exploring the space of potential clustering results by visualizing previous and current solution states, and providing choices of modalities by which the user can restrain how parameters are updated. These features help with the investigation of the quality of different clustering results (see the subsection below) in relation to the users' analytical tasks.…”

A User‐based Visual Analytics Workflow for Exploratory Model Analysis

“…Sedlmair et al [SHB∗14] provide a comprehensive survey of visual analytics tools for analyzing the parameter space of models. Example types of models used by these visual analytics tools include regression [MP13], clustering [NHM∗07, CD19, KEV∗18, SKB∗18], classification [VDEvW11, CLKP10], dimension reduction [CLL∗13, JZF∗09, NM13, AWD12, LWT∗15], and domain‐specific modeling approaches including climate models [WLSL17]. In these examples, the user directly constructs or modifies the parameters of the model through the interaction of sliders or interactive visual elements within the visualization.…”

Immersive Insights: A Hybrid Analytics System forCollaborative Exploratory Data Analysis

“…Due to the lack of a clear mapping between the axes generated by DR techniques and the original data dimensions, earlier work in data analysis has proposed various tools and techniques to guide users in exploring low-dimensional projections of data [28,48]. Methods such as rotation and isolation [33] enable the user to interactively rotate multivariate data, and statistical information can be used to structure possible visualizations of the data [6,17,27,70,76,79,80]. Since low-dimensional projections are generally lossy representations of high-dimensional data relations, researchers have introduced visual methods to convey and correct dimensionality reduction errors [7,20,45,71].…”

“…To contextualize the various assignments generated by clustering algorithms, tools such as ClustVis [56], Clustrophile [26] and ClusterVision [44] coordinate visualizations of discrete clusterings with scatterplot visualizations of dimensionality reductions. Correlation and ANOVA-based significance analyses are seamlessly integrated in the clustering process in Clustrophile 2 [17].…”