Automatic Abstraction Management in Information Visualization Systems Marcelo Campo - Ricardo Orosco* - Alfredo Teyseyre Univ. Nacional del Centro Prov. Bs. As - Fac. Ciencias Exactas ISISTAN- Grupo de Objetos y Visualización, San Martín 57, (7000) Tandil, Bs. As., Argentina E-Mail: mcampo@necsus.com.ar Ricardo.Orosco@ii.uam.es * also Universidad Autónoma de Madrid, Depto. Ingeniería Informática Campus Cantoblanco, Mod. I-V, 28049, Madrid, España Comisión Investigaciones Científicas (CIC) Prov. Buenos Aires, Argentina Abstract The construction of information visualization systems is a difficult task. However, there are few works in the provision of software architectures for visualization systems, in order to reduce this difficulty. Particularly, systems that provide a reusable support for the automatic management of the different levels of abstraction in which complex data can be observed are not found in current visualization systems. In this work, Telescope, an object- oriented architecture for visualization systems is presented. Its main goal is to provide a customizable infrastructure to develop visualization systems allowing the automatic management of the different levels of abstraction in which the visualized information can be observed. Telescope is based on the concept of abstractor objects which provides the generic behavior to support visualizations with semantic zoom capabilities controlled externally through abstraction scales. CityVis, a visualization system for city data, developed using the Telescope architecture, is also described. 1. Introduction Data visualization is a powerful tool to facilitate the analysis and understanding of complex information. Although its increasingly importance, both in research and commercial communities, the process of creating these visualizations is still quite complex. In most cases, visualizations tend to be hand-crafted, each one different from the previous, making difficult to reuse implemen- tations from one visualization to another. Several factors contribute to this difficulty: • there are a great variety of domains that can be visualized, each one with its own requirements and characteristics. • the analyzer may perform distinct types of visualization processes [4]: exploratory, analytical or descriptive. • users can perform visualizations with different goals, and/or use distinct visualization techniques according to their purposes. • there is a low reusability degree of previous visualizations, as a consequence of the hand- crafted nature of visualization systems or because their designs are oriented to particular domains. However, in spite of this difficulty, the way in which visualizations can be implemented has received relatively little attention. Most of current works on information visualization are focused on the development of new visualization techniques (e.g. [11], [19], [7], [1], [13]), or in the provision of assistance to the user in the data exploration process (e.g. [18], [14], [2]), but little research have been reported about software architectures that make simpler the construction of visualizations of complex data. Particularly, systems that provide a reusable support for the automatic management of the different levels of abstraction in which complex data can be observed are not found in current visualization systems. The management of abstraction levels is one of the main features in any data visualization. However, the construction of techniques