Searching and Browsing Text Collections with Large Category Hierarchies

ABSTRACT

Keywords

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

INTRODUCTION

Some attributes that are often associated with documents, (sometimes called meta-data) describe aspects of the external properties of the document, including author, date of creation, provenance, document length, language, and so on. Certain recent interfaces, including Envision [4] and SenseMaker [1], have focused on allowing the user to see which documents are associated with various combinations of this kind of information.

Another kind of attribute information has to do with the actual content of meaning of the documents. These attributes are often called category labels or subject codes (e.g., in bibliographic collections, such as seen in the Dewey Decimal system). Some online text collections now have associated with them very large, complex sets of category labels. For example, the Association for Computing Machinery (ACM) has associated with it a large subject hierachy consisting of approximate 1200 labels. These category labels (e.g., Hardware Description Languages, Image Processing) are assigned by authors to their journal articles in order to indicate the subject matter of the documents, averaging approximately three to five categories per document.

Once these labels are available, how should a text retrieval system provide access to a text collection that has associated with it a large hierarchy of categories? In most systems, and in many World Wide Web interfaces, the user must scroll through a list of categories to find those of interest. The user is usually not given an overview of the category space other than a top-level view of the most general labels. If the system does let the user search over the names in the category labels, the results are typically shown as an alphabetical list, with corresponding documents listed under the corresponding category labels.

Thus these approaches do not show users the context in which the category labels or are used within the documents, and make navigation of very large category hierarchies difficult. Futhermore, if documents have been assigned multiple categories, most interfaces provide no mechanism for helping the user make use of the multi-part description of the document's contents.

SEPARATING CATEGORIES FROM DOCUMENTS

To achieve the separation between category labels and documents, category labels are placed in a ConeTree [8], and after a search over free text words or category labels, retrieval results a placed in a virtual book [2]. By contrast, in most systems that present graphical hierarchies, documents are associated with each node of the category hierarchy; clicking on a node reveals the documents assigned this node. In the Cat-a-Cone, documents as associated with searches. When the user opens up a ``page'' of the retrieval results, the parts of the hierarchy that correspond to the set of categories that have been assigned to the document are shown in the ConeTree. The virtual book allows the user to ``flip'' through a set of pages, and this causes differences among documents to be animated in terms of which parts of the hierarchy remain present, which shrink away, and which sprout out.

The use of the ConeTree also enables the user to see the context in which the category labels occur. Thus the meaning of unfamiliar or ambiguous categories can be made clearer by display of their ancestors, siblings, and immediate descendants. Different subtrees of the hierarchy can be displayed at different levels of granularity depending on the familiarity to the user.

Figure 1:   The Cat-a-Cone interface applied to the Yahoo hierarchy.

SPECIFYING QUERIES

The Cat-a-Cone is not meant to be a stand-alone system, but rather part of a suite of tools each designed to help with a different aspect of the information access problem. We plan to evaluate the usefulness of our ideas on medical text using as subjects cancer patients and clinicians.

References

1

Michelle Q. Wang Baldonado and Terry Winograd. Sensemaker: An information-exploration interface supporting the contextual evolution of a user's interests. In Proceedings of the ACM SIGCHI Conference on Human Factors in Computing Systems, 1997. To appear.

2

Stuart K. Card, George G. Robertson, and William York. The webbook and the web forager: An information workspace for the world-wide web. In Proceedings of the ACM SIGCHI Conference on Human Factors in Computing Systems, Vancouver, Canada, April 1996.

3

Douglass R. Cutting, Jan O. Pedersen, David Karger, and John W. Tukey. Scatter/Gather: A cluster-based approach to browsing large document collections. In Proceedings of the 15th Annual International ACM/SIGIR Conference, pages 318-329, Copenhagen, Denmark, 1992.

4

Edward A. Fox, Deborah Hix, Lucy T. Nowell, Dennis J. Brueni, William C. Wake, Lenwwod S. Heath, and Durgesh Rao. Users, user interfaces, and objects: Envision, a digital library. Journal of the American Society for Information Science, 44(8):480-491, 1993.

5

Marti A. Hearst. Tilebars: Visualization of term distribution information in full text information access. In Proceedings of the ACM SIGCHI Conference on Human Factors in Computing Systems, Denver, CO, May 1995.

6

Marti A. Hearst and Chandu Karadi. Cat-a-cone: An interative interface for specifying searches and viewing retrieval results using a large category hierarchy. 1997. Submitted for publication.

7

William R. Hersh, David H. Hickman, Brian Haynes, and K. Ann McKibbon. A performance and failure analysis of saphire with a medline test collection. Journal of the American Medical Informatics Association, 1(1):51-60, 1994.

8

George C. Robertson, Stuart K. Card, and Jock D. MacKinlay. Information visualization using 3D interactive animation. Communications of the ACM, 36(4):56-71, 1993.

9

Padmini Srinivasan. Optimal document-indexing vocabulary for medline. Information Processing and Management, 32(5):503-514, 1996.