Exploring Search Results with Envision

ABSTRACT

Keywords

© 1997 Copyright on this material is held by the authors.

INTRODUCTION

Partly in response to these problems, information retrieval researchers have developed free text search systems based on various forms of approximate matching. These systems typically present results as text lists, ordered by how closely the documents match the query. Approximate match searchers typically return very large sets, including many works that users find irrelevant. Further, while relevant works are likely to be close to the top, they are often mixed with irrelevant works in ways that are opaque even to experienced users. Finally, since results are not in bibliographic order, users have difficulty finding known works in the list and often lose their place.

Several recent systems, including TileBars [5], InfoCrystal [9], and VIBE [8], combine free text retrieval with visualization techniques. These and most other such systems have concentrated on depicting how well documents in the result set match the query. The information visualized is unchangeable, although the displays may include zoom and other manipulations.

ENVISION SYSTEM AND DESIGN

The Envision system consists of a query server, an object-oriented database management system, a presentation server, and the Envision client. The Envision query server uses the MARIAN vector space search engine running on a test collection of 100,000 documents [3].

Design of the Envision user interface was motivated by intensive interviews with twelve potential users, all established researchers in computer and information science [4]. Beyond ready access from their offices, chief among interviewees' wishes was the ability to identify and explore patterns in the literature. Some asked for visual representations, while others wanted to see connections not visible with current tools. Thus we turned to visualization in attempting to meet user needs. Development has proceeded iteratively, with extensive usability evaluation involving a wide range of participants [7].

As shown in the figure, each document in a search results set is shown graphically as an icon in the Graphic View window, which somewhat resembles a starfield display [1]. The Item Summary window shows a textual listing of bibliographic information for documents whose Graphic View icons are selected. Additional details, including full content, are presented on demand using Mosaic and related viewers. The Graphic View supports users in making decisions about which works to examine in potentially large sets of documents. Since users' perceptual strengths vary and their decision criteria reflect their current information needs, each graphical device in the Graphic View is user-controllable to represent different document attributes at different times. Document characteristics that may be visualized include similarity to the query, publication year, document type (e.g., book journal, journal article, video), author names, and index terms. Icon characteristics used in the visualizations include placement relative to the x-axis and y-axis and an alphanumeric icon label, as well as icon size, shape, and color. Formative usability evaluation has shown Envision to be a highly usable systems. Users especially like the power and flexibility of the Graphic View Window [7].

A primary goal in design of the Envision user interface has been to allow users to explore patterns in the collection. For example, by displaying relevance on both the x-axis and y- axis, a researcher can see drop-offs in the relevance numbers and gain insight into performance of our search engine. Displaying author on the y-axis and publication year on the x-axis while using icon color to show probable relevance and icon shape to show document type, a user might determine which authors have recently published highly relevant works proceedings articles. These visualizations and others will be demonstrated and discussed in terms of user tasks supported.

ACKNOWLEDGMENTS

REFERENCES