Formative Evaluation of Near-Semantic Search Interfaces

Abstract

This article reports findings from a usability study of an experimental near-semantic search interface. We present user preferences for exploring semantic recommendations and model a new user interface for subject suggestions based on the results. The resulting catalog search engine (Deneb 2.0) is in public beta implementation, available at: http://dunatis.grainger.uiuc.edu/deneb-2.

KEYWORDS:

• semantic web
• data mining
• usability

Acknowledgments

This study was made possible through the collaboration between colleagues and campus units. The authors wish to acknowledge the Research and Publication Committee of the University of Illinois at Urbana-Champaign Library and the University of Illinois Campus Research Board, which provided support for the completion of this research. We would also like to thank Yinan Zhang, for his development work and information retrieval expertise, as well as Jennie Archer, Amanda Hatland, and Rubayya Hoque, who assisted in the recruitment, observation, and interviewing of the usability test participants.

©Jim Hahn and Chris Diaz

Notes

1. The semantic web is defined in the Encyclopedia of Library and Information Sciences as “an extension, in progress, to the World Wide Web (WWW), designed to allow software processes, in particular artificial agents, as well as human readers, to acquire, share, and reason about information” (O’Hara & Hall, 2011, p. 4663).

2. Deneb stands apart from the VuFind installation at Illinois; as an alternative layer, it is derived from a snapshot of catalog data, and has no direct connection to the ILS (Voyager). Through unique bibliographic identifiers all Deneb links point back to the VuFind catalog. Deneb 1.0 and 2.0 are powered by the open source search engine Apache Lucene (http://lucene.apache.org/).Deneb therefore could work with any ILS technology using unique bibliographic identifiers – e.g. most contemporary integrated library systems.

3. The research and development test dataset (pre-Deneb 1.0) was 33,372 titles consisting of 3556 terms (Wang et al., 2013, p. 442). The test data set expanded over the time of our grant to encompass the entirety of the catalog's 13 million volumes in both Deneb 1.0 and 2.0.

4. We use “near-semantic” in this article to denote that the research and development process deviates from traditional semantic web experimentation, particularly in the aims and uses of publishing linked open data, which was not a research goal of our study.