|
|
Data visualisation, Geographic Information Systems (GIS), knowledge-based systems, World Wide Web.
© 1997 Copyright on this material is held by the authors.
Presentation of spatially referenced data on maps is an important prerequisite for analysis of the data. Currently existing mapping software (GIS), despite of its functional power, does not fit well the needs of data exploration. It requires the user to think what presentation of her/his data will be adequate, and what GIS operations must be applied to obtain it. Besides consuming user's time, design of consistent and informative map presentations requires special cartographic knowledge that is hardly possessed by any end-user. Therefore the need in intelligent tools providing assistance to GIS users in data visualisation is widely recognised in the GIS community.
General principles of graphical presentation that should be followed in data visualisation design are described by a number of authors, Bertin [1] being the best known. Mackinlay [3] showed the feasibility of applying these general principles for automated data visualisation. Jung [2] extended Mackinlay's approach to cartographic presentation of spatially referenced data. According to the approach, a set of data components selected for visualisation is partitioned so that it becomes possible to find for each partition some graphical primitive (position, size, colour, etc.) in agreement with characteristics of its components. Then the primitives selected are combined by available composition operations. When this occurs impossible, backtracking is done to the stage of assigning primitives or partitioning.
The research prototype system IRIS presented here takes into account peculiarities of the map form of data presentation and relies upon metainformation about relationships among data components. "Building blocks" in IRIS are the traditional visualisation techniques developed in cartography [4], such as choropleth map or diagrams (bar charts, pie charts, etc.) referred to geographical objects. Each visualisation technique has specific requirements to data characteristics, such as the type of value domain or cardinality, and to relationships among data components it is applied to. For example, there are techniques that require the data components to be comparable, or additive and forming some whole, or linked by part-of relation. On the basis of metainformation, the system does not consider arbitrary partitions but unites components in one group only if they are linked by one of these relationships. The relationship determines selection of visualisation technique. This reduces the search space and increases design efficiency.
Data to be analysed in IRIS should be stored in a table, i.e. a collection of uniform records (rows) composed from fields; the latter form table columns. The data should refer to some geographical objects listed in one of the columns. Metainformation needed for intelligent data handling includes notions associated with each column, their characteristics, and relationships. For example, it may be stated that the column 17 of the data file europe.dbf contains values of the numeric attribute "population number" for the parameters "sex class"="female" and "age class"="15-64 years", and that "0-14 years", "15-64 years", and "65 years and over" constitute "whole population"; quoted are domain notions. Typical use of IRIS implies creation of applications with metainformation specified beforehand. When such information is not available, the system applies in map design only the visualisation techniques suitable for unrelated data.
To obtain map presentation of data from a table, the user needs only to select the columns s/he wishes to be shown simultaneously. The system then automatically produces maps according to the principles of map design in agreement with data characteristics and relationships.
Shown below is an example map built by IRIS. It presents three columns from a table with demographic data referring to the countries of Europe: "dominant religion", "population number, all age classes total", and "population number, 65 years and over". "Dominant religion", as an attribute with nominal value domain, is shown by colouring contours of countries; each religion is assigned a distinct colour. Population numbers (quantitative data) are shown by areas of nested squares; this diagram type is allowed by the part-of relationship between "65 years and over" and "all age classes total". Combination of painting with diagrams is one of the few composition operations allowed for maps.
In most cases IRIS generates more than one map for the same data. Contrary to the common practice in knowledge-based data visualisation, we intentionally do not rank visualisation techniques by their effectiveness. Having several presentations, the user can choose the most convenient, or explore data by comparing several maps. Besides, up to now there were no experimental results that could allow to rank visualisation techniques by another principle than accuracy of decoding separate data values. We believe that this principle is not so important for interactive computer presentations. In IRIS, when the user points with the cursor to a geographical object, the system displays exact data values referring to this object. A visualisation technique is effective for data exploration if it gives the overall view on data and helps in understanding their inherent features and relationships. We hope that the use of IRIS in experiments with different people and different data will help us in evaluating visualisation techniques from this perspective.
Besides automatic data mapping, IRIS provides some other functions useful for data exploration. One of the functions is filtering: the user specifies some restrictions on data values, and the system selects only the records satisfying the restrictions. After this all visualisations will be applied to the selected data subset. The system enables and, or, and not operations over filters that allow to specify rather complex queries.
The system can also perform calculations by an arbitrary arithmetic formula over table columns specified by the user. Calculated values are put into a new column. This column then may also be visualised, separately or together with other columns. We also develop facilities for map analysis by dynamic transformations when a map changes its appearance in response to user's actions such as switching on/off the display of particular values or intervals.
IRIS is implemented in two variants: one runs on PC under Windows, another is installed in the WWW with the interface part written in Java and the core part written in C++ and working on a UNIX-based server. The latter realisation allowed wide access to the system via the Internet. Registration of actions of people who worked with the system gave us some ideas that allowed to improve the characteristics of the system.
We consider as the primary merit of IRIS the fact that it releases the user from all routine work on map building and thus allows to focus on data exploration. The other advantage of the system is that it does not require the user to have special cartographic knowledge. Instead, this knowledge is included into the system. The system acts as a qualified cartographer creating necessary prerequisites for the user to find interesting and often unexpected facts about the data under analysis.
The goals pursued in the development of IRIS have much in common with those of the software for data mining and knowledge discovery in databases. We believe that our system could be organically integrated with such software. The synergy of automated knowledge discovery methods and visual data analysis by the user may give much more than each of the tools alone.
We thank GMD (German National Research Centre for Information Technology) that kindly invited us as guest researchers and thus gave us the opportunity to work over IRIS.
|
|