ACM Computing Surveys 28A(4), December 1996, http://www.acm.org/surveys/1996/CruzTailorable/. Copyright © 1996 by the Association for Computing Machinery, Inc. See the permissions statement below.

---

Abstract: Information visualization is emerging as one of the most important topics in human-computer interaction. Existing systems have been successful at providing automatic ways to display information using sophisticated rendering techniques (e.g., 3D, animation). We argue for tools with which users can adapt visualizations to the application at hand and to their own preferences. Ideally, such tools should combine user-defined displays with automatically-generated displays and be integrated with search mechanisms capable of accessing data in a variety of information bases (e.g., digital libraries, the WWW, databases). A combined effort of researchers in human-computer interaction and in information systems is therefore needed.

---

Introduction

• Locating the data
  This can be achieved with data mining and knowledge discovery tools [Etzi96].
• Processing the data
  This involves selecting and transforming (e.g., aggregating) hypermedia/multimedia data using query languages or other information retrieval mechanisms [CMW95].
• Presenting the retrieved information
  The information may consist of text, numerical data, video clips, and images, which need to be spatially and temporally put together [WeWi94, CMW95].

This paper focuses on information visualization from the HCI point of view, while pointing to relevant issues in information retrieval and databases. We argue for making tools available that give the end user the power to tailor the information displays. Because of the diversity of users, providing different levels of customization is another desirable goal: the idea being that some users will be able to define new visualizations from scratch, while other users will be happy to change just a few parameters to the visualization. Next we discuss several aspects that may lead to better tools for information visualization in general, and to tailorable information visualization in particular.

Information Visualization Approaches: the HCI and Database Viewpoints

In the database community, the focus on information visualization started with research in visual query languages, where the visualization of schema and/or database instances is common (for a survey, see [CCLB97]). Recently, a new generation of database systems is emerging, which tightly combine querying capabilities with visualization techniques and are information visualization systems in their own right [CCS94, Cruz92, HIL95]. These systems also allow for tailoring the displays to the users' preferences.

There is a host of research opportunities in HCI associated with database applications that access large data repositories (e.g., data mining and data warehousing), and with information retrieval (e.g., in digital libraries and the WWW) [FoPi94, FAFL95]. The research issues include providing both "effective" visualizations of large amounts of documents and carefully designed user interfaces capable of addressing the needs of a wide variety of users.

Automatic Visualization

Other approaches whose main focus is not on the layout of graphs include systems like APT [Mack86] and Sage [RKMG94]. These systems have focused on the automatic generation of visualizations given the semantics of the data (e.g., functional dependencies in a relational database in APT).

An comprehensive approach is proposed by [Rao et al.95], where 3D displays (e.g., the "Document Lens", the "Butterfly Application", and the "Hyperbolic Tree Browser") and search mechanisms are integrated in a single environment for accessing digital libraries.

Declarative Approaches

An issue of practical importance concerns the efficiency of rendering a picture. Techniques that rely on optimization techniques (e.g., simulated annealing or general constraint solvers as in [Mark91, CrTw96]) are usually inefficient. For large data sets, such inefficiency can become an issue especially when the displays are to be generated as the users browse the data. Therefore, the identification of layouts for which a constraint solver exists that matches the best known algorithm would be an important accomplishment. Preliminary results show that a significant class of graphs can be rendered in linear time [CrGa94].

Incorporating an automatic approach to data visualization with a user-defined approach is an interesting and practical problem, for example when the user-defined visualization is underconstrained. Ideally, an automatic layout facility could supplement the user-defined visualization.

The users' preferences could also provide invaluable information on ways to automatically perform the display of the data: how can such knowledge be gathered to automate the display of information for users who have previously defined displays of "similar" data? How can preferences of groups of users be integrated and taken into account for future displays?

Expressiveness and Readability of Visualizations

A related issue concerns the readability of the displays, and the need to understand how users interpret information displays. In graph drawing, the readability is associated with (often conflicting) aesthetic criteria such as the minimization of edge crossings and of the area of the graph, and the maximization of symmetries. A better display and improved readability can also be achieved by using 3D graph visualization techniques [RMC91, GaTa96, CrTw96, CEH96].

Readability studies have been performed for 2D graph drawing [PCJ96], but more such studies are needed, including studies for 3D visualizations. Automatic visualization approaches can benefit in obvious ways from a better understanding of readability and expressiveness issues. Incorporating such knowledge in tailorable visualization tools is, however, a more difficult task: how can "inappropriate" visualization choices be detected and how can useful feedback be provided to the user?

New Applications: the Challenges

The platform issue is particularly relevant for Internet and Intranet applications. In this regard, consideration should be given to platform-independent programming languages such as Java, that can in addition support algorithm animation, e.g., for education [BCLT86].

In distributed environments, the specification of the visualization should take into account accessibility conditions to the different information bases, e.g., to take into account different times to receive the information from the different locations. Ways to "dynamically" specify the layout should therefore be supported. A somewhat similar situation occurs when loading a Web page, where pictures may fail to appear or first appear incomplete. Other time-related issues occur when the information itself varies over time (as with audio or video) [Zell95].

New standards for Web documents should include an object-oriented approach to the representation of hyperlinks and Web documents. In this way, it is possible to specify the selective browsing, querying, and visualization of electronic documents that belong to a class (and its subclasses) and/or are connected by hyperlinks of a class (and its subclasses). The current indiscriminate browsing (e.g., of Web documents) can be supported by allowing for documents of all classes to be accessed or hyperlinks of all classes to be traversed. Metadata information can be equally helpful to direct the querying and browsing of Web documents and to facilitate their visualization.

Acknowledgements

References