Visual Analytics methods to steer the subspace clustering process


Dr. Enrico Bertini, Konstanz
gemeinsam mit Prof. Dr. Oliver Deussen, Konstanz
Universität Konstanz
Fachbereich Informatik und Informationswissenschaft
Arbeitsgruppe Computergrafik und Medieninformatik Konstanz

Prof. Dr. Thomas Seidl, Aachen
Rheinisch-Westfälische Technische Hochschule Aachen
Lehrstuhl für Informatik IX - Datenmanagement und Exploration


Visual Analytics methods to steer the subspace clustering process


The main goal of this proposed project is the tight integration of visual analytics into the process of subspace cluster analysis to support the domain scientists’ exploration processes through a highly interactive immersive visualization. The considered databases from different fields of scientific and engineering research are usually very large and high dimensional. An approach
solely based on automated subspace cluster analysis is rarely appropriate to provide the necessary insights into the various patterns, which are usually hidden by the huge amount and the heterogeneity of the data. Appropriate visualization techniques could not only help in monitoring the clustering process but, with special mining techniques, they also enable the domain expert to guide and even to steer the subspace clustering process to reveal the patterns of interest. To this goal we envision a concept that combines scalable subspace clustering algorithms and interactive scalable visual exploration techniques. This work will include the tasks of (1) comparative
visualization and feedback guided computation of multiple alternative clusterings; (2) design of anytime subspace clustering algorithms, visualization of preliminary clustering results, intuitive annotation of these results and insertion
of feedback into the algorithms; (3) methods for incremental adaptation of the analysis to data modifications. To the best of our knowledge the whole
idea of using a visual analytics approach for steering clustering models is totally novel. The research we want to carry out in this project has the potential to open a new line of research that is not necessarily limited to subspace
clustering but applies to any other modeling technique in which the involvement of end-users in the model building process can compensate for the limits the necessary heuristics introduce in the process.