Visual Information Seeking: Tight Coupling of Dynamic Query Filters with Starfield Displays

Christopher Ahlberg and Ben Shneiderman

Summary/Analysis by Laura Slaughter

Summary:

In this paper, Ahlberg and Shneiderman define three principles for building visual information seeking interfaces. They were dynamic query filters, starfield displays and tight coupling. Dynamic query filters were described as tools such as sliders or buttons that can be used to change the parameters of a query. Starfield displays are 2D scatterplots used to present information visually, providing overviews and "zoomed in" previews of data. Tight coupling is a term that refers to the linked interactions of components within an interface. These three principles are an extension of prior work concerning direct manipulation, a term that is used to encapsulate interface mechanisms that use 1) visual representations, 2) selection of objects without typing, 3) rapid but reversible actions and 4) immediate as well as continuous display of results. A prototype system called the FilmFinder was used to demonstrate how these principles might be employed to allow users to browse a large set of data.

The authors created the FilmFinder so they could test the proposed principles. In addition, the FilmFinder provides a concrete example of the outlined principles so that the reader can better understand the definitions given in opening sections of the paper. The FilmFinder was designed as a tool for exploring a film database and is intended as a system for "making the decision process easier for groups of viewers." Alphasliders, toggle buttons and checkboxes are used to refine database queries. They illustrate dynamic filtering of the data by allowing users to input a query through direct manipulation. Titles, actors, actresses, directors, film duration, and genre parameters can be selected and corresponding result sets are then returned in the starfield display. The starfield display is a scatterplot of colored spots to represent movies in the database. Its Y-axis is a measure of popularity and the X-axis represents date (year) of movie release. The overview of the data is continuously available. The preview screen that lists all the data about a particular movie can be obtained by clicking on a colored spot. Zooming to show a close up view of a region of the starfield display can be accomplished by reducing the number of values on the axes. Tight coupling occurs between the interrelated starfield display and dynamic query filter components. When the user manipulates the dynamic query filters, the starfield display rapidly reflects those changes.

Contribution of the paper to HCI:

This paper made a significant contribution to the field of HCI and is often cited in the literature related to information visualization. The paper explained several principles that had been used prior to its publication in other applications (such as tight coupling in word processors) but that had not been formally summarized. Further, it related the benefits of these principles to interfaces for visual information seeking. Work in information visualization is still very new and this paper laid the groundwork in 1993-4 for research that is taking place today. The authors established that use of the principles support browsing of databases by using "rapid filtering to reduce results sets, progressive refinement of search parameters, continuous reformulation of goals and visual scanning to identify results." The paper furthers work in HCI because it supplies the community with a framework for the design of good visual information seeking interfaces.

Questions and troubling aspects of the paper:

  1. The principles are not actually defined in a way that provides a clear understanding. It is only through the "case study-like" examples that the reader learns what the terms mean. The "several aspects" of tight coupling are particularly confusing. The reader might first think that these are part of the definition, however, the aspects are not specific to tight coupling since they apply to the other principles as well as direct manipulation.
  2. The authors claim at the beginning of the paper that while studying visual information systems for both novices and expert users, they "have found several user interface design principles that consistently lead to high levels of satisfaction." Yet, the FilmFinder itself lacked usability tests or controlled studies that provide proof.
  3. The starfield display definition included a statement about the usefulness of a "scatterplot with ordinal axes." Popularity may have been ordinal data but it isn't presented that way. Time and popularity on the FilmFinder appear to be continuous. If the authors wanted to demonstrate the starfield display as compared to a scatterplot, maybe it would have been more interesting to use genre and ratings (nominal scale, discrete data) on the axes.
  4. Also related to the axes on the FilmFinder--Why did the authors select popularity for the axes? It is possible that the user isn't interested in other people's "popularity" measures. How do does the user know that these people have the same taste that they do? Did the authors consider giving the user the ability to change the axes to get another view of the information space?
  5. The authors mention that the FilmFinder is a tool for "groups" but they don't relate the principles to collaborative/cooperative visual information seeking systems.
  6. The authors state that the principles can be used for a "database of documents". Some readers may infer that these principles can be used to build interfaces for text IR systems. However, they have only provided principles relating to database system queries. Text IR problems are a whole different set of issues and it is dangerous to imply that these principles would be equally successful there as well.
  7. The "Dynamic Query Filters" section discusses that "query components acted as a filter, reducing the number of items left in the result set" for the HomeFinder. The reader can then assume that there were only AND queries with no OR function since an OR would increase the size of the result set. It would have been useful to know how users accomplished OR queries. Perhaps users would benefit from a new principle that allowed them to better understand OR logic.
  8. The authors barely mention human cognitive and perceptual capabilities for processing visual information. The work in this paper could have been supported with connections to the psychological aspects of visual displays of information and how they might be used to support decision making.
Spotfire