Interactive Poster: Exploring OLAP Aggregates with Hierarchical Visualization Techniques Svetlana Mansmann * Marc H. Scholl † Daniel A. Keim ‡ Florian Mansmann § University of Konstanz Abstract Analysts interact with data cubes in a predominantly “drill-down” fashion, i.e. from coarse grained aggregates towards the desired level of detail. Hierarchical visualization techniques offer an ad- equate structure for mapping the logic of the exploration due to preserving the overview of the entire interaction and providing the details of every drill-down step. We present a novel user interface for exploring complex multidimensional data in an OLAP (online analytical processing) environment. Users navigate in dimensional hierarchies via a schema-based data browser. Query results are pre- sented in form of enhanced decomposition trees. We propose mul- tiple tree layouts and embedded visualization techniques optimized to satisfy various criteria. Keywords: hierarchical visualization, OLAP, explorative inter- face. Index Terms: E.1 [Data]: Data Structures—; H.5.2 [Informa- tion Interfaces and Presentation]: User Interfaces—Graphical user interfaces (GUI) 1 Introduction OLAP systems employ the multidimensional data model to struc- ture “raw” data into cubes in which analytical values, referred to as measures, are uniquely determined by descriptive values drawn from a set of dimensions. The desired view of the data for further analysis and exploration is retrieved via a visual interface. Typi- cally, users proceed from a high level overview towards finer gran- ularity via stepwise decomposition of aggregates along dimensions of interest. Insights obtained at one stage of exploration often influ- ence the subsequent interaction. Hierarchical visualization techniques are an adequate form of presenting the results of iterative decomposition due to preserving the overview of the entire interaction and providing the details of every drill-down step. A decomposition tree is obtained by plac- ing an aggregate value in the root node and recursively splitting the values along selected dimensions to show the constituent sub- aggregates as the aggregate’s child nodes. ProClarity was the first to incorporate a Decomposition Tree [3] visualization into an OLAP interface. An OLAP web client Report Portal 2.1 [4], recently released by XMLA Consulting, enhances visual decomposition by introducing BarChart Tree and PieChart Tree. Rather than displaying plain numbers in the nodes, these tech- niques arrange the child sub-aggregates of a value into a chart, as illustrated in Figure1 (top). The existing decomposition techniques expand a single value per interaction, have rather poorly formatted results and are wasteful in terms of display utilization. * e-mail: [email protected] † e-mail:[email protected] ‡ e-mail:[email protected] § e-mail:[email protected] SECTION FACULTY DEPARTMENT 2 6.59 5.4 2 7.56 10.04 9.43 12.63 12.41 8.76 7.72 13.43 German Studies Slavic Studies Oriental Studies Romanic Studies Brit. & Amer. Studies Literature, Arts & Media 17.46 4.2 10.7 1 1 2 3. 1 Mathematics Computer Science Physics Chemistry Biology Psychology Inf. Management Life Science Bioinformatics Math. Finance Economy Politics Management Economic Science History Philosophy Sociology SECTION FACULTY DEPARTMENT Natural Sciences 53.86 Interdisci 17.48 Econ.&Politics 29.83 Linguistics 12.52 Social Sciences 32.41 Humanities 44.93 Natural & Econ. Sciences 101.17 Figure 1: Hierarchical decomposition techniques allow analysts to in- teractively ”drill down” in OLAP data and explore the aggregates at various granularity levels. Both visualizations show the university’s expenditures disaggregated along section, faculty, and department. Top: A bar-chart tree with uniform scaling within each level is espe- cially helpful for visually comparing the values of the same granularity. Bottom: Space-filling bar-chart tree is more compact; aggregates are mapped to bars of equal height providing correct visual comparison of values throughout the entire hierarchy. In a previous work [6] we described a framework for generating user-defined visual hierarchies from OLAP cubes using schema- based navigation. An approach to navigating in complex and irreg- ular dimensional hierarchies is provided in [2]. 2 Hierarchical Decomposition Our contribution is twofold: 1) we have designed a framework for efficient and intuitive exploration of OLAP data using visual hier- archical decomposition, and 2) we propose multiple visualization techniques and interaction features optimized along various criteria (e.g., visual scalability, interpretability, outlier recognition). Spe- cial effort was put into simplifying the process of generating a user- defined hierarchy by properly interfacing the data via an enhanced OLAP browser. Data cube’s schema is used for specifying the de- composition axes while dimensional data is accessed for filtering the subset to display.
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Interactive Poster: Exploring OLAP Aggregateswith Hierarchical Visualization Techniques
Svetlana Mansmann∗ Marc H. Scholl† Daniel A. Keim‡ Florian Mansmann§
University of Konstanz
Abstract
Analysts interact with data cubes in a predominantly “drill-down”fashion, i.e. from coarse grained aggregates towards the desiredlevel of detail. Hierarchical visualization techniques offer an ad-equate structure for mapping the logic of the exploration due topreserving the overview of the entire interaction and providing thedetails of every drill-down step. We present a novel user interfacefor exploring complex multidimensional data in an OLAP (onlineanalytical processing) environment. Users navigate in dimensionalhierarchies via a schema-based data browser. Query results are pre-sented in form of enhanced decomposition trees. We propose mul-tiple tree layouts and embedded visualization techniques optimizedto satisfy various criteria.
Keywords: hierarchical visualization, OLAP, explorative inter-face.
Index Terms: E.1 [Data]: Data Structures—; H.5.2 [Informa-tion Interfaces and Presentation]: User Interfaces—Graphical userinterfaces (GUI)
1 Introduction
OLAP systems employ the multidimensional data model to struc-ture “raw” data into cubes in which analytical values, referred toas measures, are uniquely determined by descriptive values drawnfrom a set of dimensions. The desired view of the data for furtheranalysis and exploration is retrieved via a visual interface. Typi-cally, users proceed from a high level overview towards finer gran-ularity via stepwise decomposition of aggregates along dimensionsof interest. Insights obtained at one stage of exploration often influ-ence the subsequent interaction.Hierarchical visualization techniques are an adequate form of
presenting the results of iterative decomposition due to preservingthe overview of the entire interaction and providing the details ofevery drill-down step. A decomposition tree is obtained by plac-ing an aggregate value in the root node and recursively splittingthe values along selected dimensions to show the constituent sub-aggregates as the aggregate’s child nodes.ProClarity was the first to incorporate a Decomposition Tree [3]
visualization into an OLAP interface. An OLAP web client ReportPortal 2.1 [4], recently released by XMLA Consulting, enhancesvisual decomposition by introducing BarChart Tree and PieChartTree. Rather than displaying plain numbers in the nodes, these tech-niques arrange the child sub-aggregates of a value into a chart, asillustrated in Figure1 (top). The existing decomposition techniquesexpand a single value per interaction, have rather poorly formattedresults and are wasteful in terms of display utilization.
∗e-mail: [email protected]†e-mail:[email protected]‡e-mail:[email protected]§e-mail:[email protected]
SECTION
FACULTY
DEPARTMENT
26.595.427.5610.049.4312.6312.41 8.767.7213.43
German Studies
Slavic Studies
Oriental Studies
Romanic Studies
Brit. & Amer. Studies
Literature, Arts & Media
17.46 4.210.7 1 12 3. 1
Mathematics
Computer Science
Physics
Chemistry
Biology
Psychology
Inf. Management
Life Science
Bioinformatics
Math. Finance Economy
Politics
Management
Economic Science
History
Philosophy
Sociology
SECTION
FACULTY
DEPARTMENT
Natural Sciences
53.86
Interdisci
17.48
Econ.&Politics
29.83
Linguistics
12.52
Social Sciences
32.41
Humanities
44.93
Natural & Econ. Sciences
101.17
Figure 1: Hierarchical decomposition techniques allow analysts to in-teractively ”drill down” in OLAP data and explore the aggregates atvarious granularity levels. Both visualizations show the university’sexpenditures disaggregated along section, faculty, and department.Top: A bar-chart tree with uniform scaling within each level is espe-cially helpful for visually comparing the values of the same granularity.Bottom: Space-filling bar-chart tree is more compact; aggregates aremapped to bars of equal height providing correct visual comparisonof values throughout the entire hierarchy.
In a previous work [6] we described a framework for generatinguser-defined visual hierarchies from OLAP cubes using schema-based navigation. An approach to navigating in complex and irreg-ular dimensional hierarchies is provided in [2].
2 Hierarchical Decomposition
Our contribution is twofold: 1) we have designed a framework forefficient and intuitive exploration of OLAP data using visual hier-archical decomposition, and 2) we propose multiple visualizationtechniques and interaction features optimized along various criteria(e.g., visual scalability, interpretability, outlier recognition). Spe-cial effort was put into simplifying the process of generating a user-defined hierarchy by properly interfacing the data via an enhancedOLAP browser. Data cube’s schema is used for specifying the de-composition axes while dimensional data is accessed for filteringthe subset to display.
show
hide
(a) Schema-based navigation hierarchy (b) Area-preserving bar-chart tree
619.97
Mathematics
1209.14
Computer Science
3120.02
Physics
4387.9
Chemistry
2881.27
Biology
502.35
Psychology
902.39
Inf. Management
3538.3
Life Science
1622.73
Bioinformatics
537.44
Math. Finance Economy
474.15
Politics
569.1
Management
611.28
Economic Science
310.52
History
214.9
Philosophy
429.34
Sociology
264.4
German Studies
124.51
Slavic Studies
456.9
Oriental Studies
327.82
Romanic Studies
490.3
Brit. & Amer. Studies
230.3
Literature, Arts & Media
1673.31
Interdisciplinary
2120.1
Natural Sciences
551.51
Econ.&Politics
668.53
Social Sciences
315.66
Linguistics
492.1
Humanities
1448.3
Natural & Econ. Sciences
970.2
ALL
5000
0
3000
1000
2000
4000
(c) Multiscale HeatMap with average values
Figure 2: Data browser for querying OLAP cubes and visualization of query results as core elements of the explorative framework
The query is specified interactively by selecting a measure toaggregate upon and an aggregation function to apply, by choos-ing the dimensions to “drill down” and the values to set as filters.We pursue a schema-based navigation metaphor by representing anOLAP cube as a container of dimensions and measures, with eachhierarchical dimension as a recursive nesting of its granularities inascending order of detail. Figure 2(a) shows the navigational treeof a cube with an expanded dimensional hierarchy of Period andon-demand data display for level Calendar Year. Having selectedthe measure and the aggregate function, the user simply drags anydimension folder into the visualization area to render a new level inthe decomposition tree.Since standard visualization metaphors are poorly scalable on
large data sets, we experimented with more compact layouts forboth the trees and the nodes. Usefulness of any particular techniquedepends on various factors, such as user’s preference, available dis-play area, chosen aggregate function, data volume etc.Classical bar-chart tree from Figure1 (left) uses aesthetic tree
layout and applies the same direction to the chart bars (e.g., verticalbars in a vertical tree). In such trees, all nodes are nicely aligned forvisual comparability along the sibling as well as along the ancestoraxis. However, different scaling at each level, unavoidable for sav-ing display area, may be misleading for interpreting the visualiza-tion. A possible solution are “area-aware” bars, as demonstrated inFigure 2(b): the area of any value’s bar equals the total area of thebars of its child subtotals. Therefore, all values throughout the treeare comparable via the area of their bars.Figure 2(c) demonstrates a hierarchical HeatMap for decompos-
ing the measure’s average value. The bottom-level nodes are rep-resented as cells in an array; higher level nodes are shaped as rect-angles spanning the width of of their subtrees, similarly to the mul-tiscale matrix pattern for pivot table visualization presented in [5].A linear colormap (from light yellow to dark red) with graduallyincreasing color intensity helps to see the overall behavior of theaverage and to immediate identify outliers. Sliders at the poles ofthe scale can be used to dynamically adjust the value range of thecolormap. We also experimented with radial tree and chart layoutsand found interesting scenarios for their application.
3 Conclusion
Our proposed visual framework empowers analysts to exploreOLAP aggregates along arbitrary dimensional paths using hierar-chical visualization techniques. Decomposition trees are generatedwith a few mouse clicks and support “speed-of-thought” analysisby revealing a multiscale view of the data. By implementing vari-ous hierarchical layouts, visual metaphors, and interaction optionswe account for a variety of tasks, data patterns and user preferences.We plan to further improve the scalability of hierarchical decompo-sition by applying pixel based techniques such as proposed in [1].
Acknowledgement
This work was partially funded by the German Research Founda-tion (DFG) under grant GK-1042, Explorative Analysis and Visual-ization of Large Information Spaces, University of Konstanz. Spe-cial thanks to students Roman Radle and Andreas Weiler for theimplementation of the prototype.
References
[1] D. A. Keim, C. Panse, J. Schneidewind, M. Sips, M. C. Hao, andU. Dayal. Pushing the limit in visual data exploration: Techniques andapplications. In KI 2003, Proceedings of 26th Annual German Confer-ence on Artificial Intelligence, pages 37–51, 2003.
[2] S. Mansmann and M. H. Scholl. Extending visual OLAP for handlingirregular dimensional hierarchies. In DaWaK 2006, Proceedings of 8thInternational Conference on Data Warehousing and Knowledge Dis-covery, 2006.
[3] Proclarity Analytics 6, 2006. Online: http://www.proclarity.com/products/proclarity analytics 6.asp.
[4] Report Portal: Zero-footprint olap web client solution. XMLA Consult-ing, 2006. Online: http://www.reportportal.com.
[5] Chris Stolte, Diane Tang, and Pat Hanrahan. Multiscale visualizationusing data cubes. IEEE Trans. on Visualization and Computer Graph-ics, 9(2):176–187, 2003.
[6] S. Vinnik and F. Mansmann. From analysis to interactive exploration:Building visual hierarchies fromOLAP cubes. In EDBT 2006, Proceed-ings of 10th International Conference on Extending Database Technol-ogy, pages 496–514, 2006.
Related Documents
