Supporting Interaction with Distributed and 
Heterogeneous Information Resources

Soyeon Park
Dissertation Committee:
Professor Nicholas, J. Belkin, Chair
Professor Daniel, O. O'Connor
Professor Jose Perez-Carballo
Professor James Allan, Outside member

Abstract

This dissertation has addressed a crucial issue in the distributed information environment: how to support effective interaction of users with heterogeneous and distributed information resources. In particular, this study compared usability, performance, and searching behaviors in systems which implement interaction with multiple databases through a common interface and with multiple databases as if they were one (integrated interaction). This study represents an empirical testing of the prevalent assumption of the information retrieval (IR) literature, in particular, the database merging literature, that integrated access to multiple and distributed information resources is the appropriate mode to offer.

An experiment was conducted to compare two levels of interaction, the common interaction and the integrated interaction, in the Text REtrieval Conference (TREC) environment. The data for this study came from a variety of sources: transaction logs, thinking-aloud protocols, post-search questionnaires, demographic questionnaires, exit questionnaires, and exit interviews. Twenty-eight volunteers were recruited from the graduate students of School of Communication, Information, & Library Studies at Rutgers University. Overall, it appears that for this population, interacting with different databases through a common interface is preferable on all grounds to interacting with databases through an integrated interface. These results suggest that the general assumption that an integrated interaction is best needs at least to be revisited.

The findings from this study suggest that for digital library purposes, it is important not just to integrate the search results, but also to characterize different databases to support user choice for integration, rather than relying on blind system integration. The results of this study also suggest that it is important to allow for more user control in various ways in the distributed environment, whether it is a control over database selection or control over relevance feedback.

There are some limitations of this study regarding the nature of sampling method, sample size, the artificial environment, and the choice of databases. In spite of these limitations, the results of this study contribute to the field of information science by bringing insights into understanding the role of interaction in the distributed information environment and by suggesting system features in digital libraries.