SBU CSE 591 - A Flexible Approach for Visual Data Mining

Unformatted text preview:

A Flexible Approach for Visual Data MiningMatthias Kreuseler and Heidrun SchumannAbstractÐThe exploration of heterogenous information spaces requires suitable mining methods as well as effective visual interfaces.Most of the existing systems concentrate either on mining algorithms or on visualization techniques. This paper describes a flexibleframework for Visual Data Mining which combines analytical and visual methods to achieve a better understanding of the informationspace. We provide several preprocessing methods for unstructured information spaces such as a flexible hierarchy generation withuser controlled refinement. Moreover, we develop new visualization techniques including an intuitive Focus+Context technique tovisualize complex hierarchical graphs. A special feature of our system is a new paradigm for visualizing information structures withintheir frame of reference.Index TermsÐInformation visualization, multidimenisional information modeling, hierarchies, focus+context techniques, clustering,maps, information analysis.æ1INTRODUCTIONEXPLORATION of complex information spaces has becomeone of the ªhot topicsº in many research fields,including computer graphics, data mining, pattern recogni-tion, and learning, and other areas of statistics, as well asdata bases and data warehousing. A variety of novel miningtechniques, visualization paradigms, and frameworks havebeen developed in recent years. Nevertheless, extractinguseful knowledge or models from observed data is still acomplicated nontrivial process.In this context, visualization offers a powerful means ofanalysis that can help to uncover patterns and trendshidden in unknown data. Additionally, visualizationprovides a natural method of integrating multiple data setsand has been proven to be reliable and effective across anumber of application domains. Still, visual methodscannot entirely replace analytic nonvisual mining algo-rithms. Rather, it is useful to combine multiple methodsduring data exploration processes [31].The new area of visual data mining focuses on thiscombination of visual and nonvisual techniques as well ason integrating the user in the exploration process. Integrat-ing visual and nonvisual methods in order to support avariety of exploration tasks, such as identifying patterns inlarge unstructured heterogeneous information or display-ing information context (e.g., frame of spatial or domainreferences), requires sophisticated mining, visualizationand interaction techniques. This carries over entirely newqualities of problems. Some of the most important ones canbe summarized as follows:. Extracting patterns and controlling the mining: Theexploration of large unstructured information spacesrequires information preprocessing. In this regardªfiltering out uninteresting itemsº and mergingsimilar objects into groups are necessary in orderto reveal hidden patterns. Suitable metrics have to beapplied for obtaining similarities and structures inhigh-dimensional feature space. Furthermore, thedegree of abstraction has to be controlled interac-tively in order to supervise and steer the search forpatterns during the mining process. This interactionis of outstanding importance to support explorationsat arbitrary levels of detail.. Visualizing information sets: The success of visual dataanalysis depends very much on its ability to supporta variety of exploration tasks such as overview,zoom in on items of interest or details on demand.Different visualization methods are required forrevealing information structure and informationcontents such as attribute values. Furthermore, novelinteraction techniques are needed for controlling thedegree of abstraction within visual representationsand for providing navigational aids in informationspace.. Visualizing the frame of reference: Effective explora-tions of spatially referenced information (e.g., healthdata in certain areas) require the combination of anadequate display of the spatial frame of referencewith the visualization of complex informationstructures. It is necessary to find an appropriatemapping between information and frame of refer-ence. In particular, we address the problem ofdisplaying complex graphs over geographical maps,a problem that has not been widely studied.Ankerst [4] classifies current visual data miningapproaches into three categories. Methods of the first groupapply visualization techniques independent of data miningalgorithms. The second group uses visualization in order torepresent patterns and results from mining algorithmsgraphically. The third category tightly integrates bothmining and visualization algorithms in such a way thatintermediate steps of the mining algorithms can bevisualized. Furthermore, this tight integration allows usersIEEE TRANSACTIONS ON VISUALIZATION AND COMPUTER GRAPHICS, VOL. 8, NO. 1, JANUARY-MARCH 2002 39. The authors are with the UniversitaÈt Rostock, FB Informatik, PostBox 999,18051 Rostock, Germany.E-mail: {mkreusel, schumann}@informatik.uni-rostock.de.Manuscript received 12 apr. 2001; accepted 10 July 2001.For information on obtaining reprints of this article, please send e-mail to:[email protected], and reference IEEECS Log Number 114504.1077-2626/02/$17.00 ß 2002 IEEEto control and steer the mining process directly based on thegiven visual feedback.A variety of visualization methods which have beendeveloped in different domains can be classified into thefirst group referring to the classification given above.Among these are techniques for visualizing multidimen-sional information. These methods try to map correlationsof objects in high-dimensional information space to spatialcorrelations in a 2D or 3D presentation space. Among theseare approaches like IVORY [10], VR-VIBE [6], and Narcissus[12], which exploit spring models to place objects accordingto their similarities, whereby similar objects are placedspatially close together. Other systems, like Lyberworld [11]and SPIRE [33], use different visual metaphors likeRelevance Spaces [11], Information Galaxies, or Themes-capes [33] in order to visualize document collections orresults from data base retrieval. FOCUS [25] is an inter-active table viewer which supports the exploration ofcomplex object-attribute tables by a combination of afocus+context technique, a hierarchical outliner for largeattribute sets and a general easy-to-use dynamic querymechanism.Other visual interfaces have been developed for visualiz-ing and interacting with hierarchies, like Cone


View Full Document
Download A Flexible Approach for Visual Data Mining
Our administrator received your request to download this document. We will send you the file to your email shortly.
Loading Unlocking...
Login

Join to view A Flexible Approach for Visual Data Mining and access 3M+ class-specific study document.

or
We will never post anything without your permission.
Don't have an account?
Sign Up

Join to view A Flexible Approach for Visual Data Mining 2 2 and access 3M+ class-specific study document.

or

By creating an account you agree to our Privacy Policy and Terms Of Use

Already a member?