Semantic Components: A Model for Enhancing Retrieval of Domain- Specific Information

Abstract

Semantic component searching allows a user to refine a topical search by indicating a preference for documents containing specific semantic components or by indicating terms that should appear in specific semantic components. 2We investigate four aspects of semantic components in this research. First, we describe lessons learned from using two methods for developing schemas in two domains. Second, we demonstrate use of semantic components to express domainspecific concepts and relationships by mapping a published taxonomy of questions asked by family practice physicians to the semantic component schemas for two document collections about medical care. Third, we report the results of a user study, showing that manual semantic component indexing is comparable to manual keyword indexing with respect to time and perceived difficulty and suggesting that semantic component indexing may be more accurate and consistent than manual keyword indexing. Fourth, we report the results of an interactive searching study, demonstrating the ability of semantic components to enhance search results compared to a baseline system without semantic components.In addition, we contribute a formal description of the semantic components model, a prototype implementation of semantic component indexing software, and a prototype implementation adding semantic components to an existing commercial search engine.Finally, we analyze metrics for evaluating instances of semantic component indexing and keyword indexing and illustrate use of a session-based metric for evaluating multiple-query search sessions.

Using semantic components to search for domain-specific documents: An evaluation from the system perspective and the user perspective

“…We found that the cost of manual semantic component indexing (with respect to time and intellectual effort) is similar to manual keyword indexing. We also found that the accuracy and consistency, while not directly comparable, appeared to be as good or better than that of keyword indexing [11]. In settings where manual keyword indexing is feasible, such as sundhed.dk, the quality level of semantic component indexing we used in this study is practical.…”

“…In other work [11], we discuss schema creation in more detail, comparing semantic component schemas to other indexing and classification structures, and discuss an alternative approach to schema creation that focuses primarily on the known purposes for the document collection and begins by identifying any existing document types or templates. In that work we also present data from an experiment that assesses indexing consistency when multiple indexers use a semantic component schema, similar to the one used in this study, to index a selection of documents from sundhed.dk.…”

“…In Section 4.1.4 we describe how we developed the semantic component schema used in this study. In other work, we investigated the use of semantic components to express information needs [7] and discussed the methods and challenges for developing semantic component schemas as knowledge representation structures for particular document collections [11].…”

“…We consulted physician users of the collection and also sundhed.dk employees who are involved in the indexing and editorial processes (and who have extensive contact with users) about how users were likely to use semantic components. Finally, we adjusted some of the labels and descriptions following a related study examining how indexers apply semantic components [11]. Fig.…”

“…We recruited indexers who provide keyword indexing for an existing digital library and taught them, in a few hours, how to apply semantic component indexing. In other work [11], we compared semantic component indexing to keyword indexing. We found that the cost of manual semantic component indexing (with respect to time and intellectual effort) is similar to manual keyword indexing.…”