Amy S. Jackson, Myung-Ja Han, Kurt Groetsch, Megan Mustafoff, Timothy W. Cole (2008). “Dublin Core Metadata Harvested through OAI-PMH” in Journal of Library Metadata. Vol. 8 no.1. (Preprint) Dublin Core Metadata Harvested through OIA-PMH Amy S. Jackson, Myung-Ja Han, Kurt Groetsch, Megan Mustafoff, Timothy W. Cole Abstract The introduction in 2001 of the Open Archives Initiative Protocol for Metadata Harvesting (OAI-PMH) increased interest in and awareness of metadata quality issues relevant to digital library interoperability and the use of harvested metadata to build "union catalogs" of digital information resources. Practitioners have offered wide-ranging advice to metadata authors and have suggested metrics useful for measuring the quality of shareable metadata. Is there evidence of changes in metadata practice in response to such advice and/or as a result of an increased awareness of the importance of metadata interoperability? This paper examines metadata records harvested over a six-year period by the University of Illinois at Urbana Champaign, and reports on quantitative and qualitative analyses of changes observed over time in shareable metadata quality. Introduction The importance of descriptive practice is not a new theme in the library domain; however, the wide-spread adoption of the Open Archives Initiative Protocol for Metadata Harvesting (OAI-PMH) and the Dublin Core Metadata scheme has led digital library practitioners to examine the characteristics of shareable, non-MARC descriptive metadata Dublin Core Metadata Page 1 of 24records. The IMLS/NISO Framework of Guidance for Building Good Digital Collections, first published in 2001, emphasizes the importance of disseminating descriptive metadata that supports interoperability. Following the publication of this document, additional concrete advice on how to create metadata well suited for sharing has been offered in several venues (Digital Library Forum and the National Science Digital Library, 2005), (Elings & Waibel, 2007), (Hutt & Riley, 2005), (Shreeves, Riley & Milewicz, 2006a), (Zeng & Chang, 2006), (Dushay & Hillman, 2003). Less frequently discussed, however, is how institutions are implementing Dublin Core in practice (Ward, 2004). The following article discusses quantitative and qualitative observations of Dublin Core metadata records harvested by two cultural heritage service providers at the University of Illinois at Urbana Champaign (UIUC). The examination focuses on changes in metadata practices over time, as well as observations of inaccurate and inconsistent mappings to Dublin Core. Researchers originally hoped to find indications of metadata becoming more shareable as digital projects mature, but findings did not support this hypothesis. UIUC Metadata Portals UIUC provides access to descriptive metadata harvested with OIA-PMH through several portals, including the Institute of Museum and Library Services Digital Collections and Content Project (IMLS DCC), located at http://imlsdcc.grainger.uiuc.edu, and the Committee on Institutional Cooperation (CIC) Metadata Portal, located at http://cicharvest.grainger.uiuc.edu. The IMLS DCC portal harvests metadata from cultural heritage projects funded by the Institute of Museum and Library Services Dublin Core Metadata Page 2 of 24(IMLS). Eighty-five percent of the records in this portal represent images and fourteen percent represent texts. The IMLS DCC project staff interacted with several data-providers regarding technical specifications and administrative information, but gave relatively little feedback to individual metadata providers regarding metadata quality. The project allowed for general presentations and publications stressing the importance of shareable metadata quality, including presentations at IMLS WebWise Conferences (Cole & Shreeves, 2004), and publications in other venues (Shreeves, Riley & Milewicz, 2006a). Conversations with data providers regarding mapping best practices were not within the scope of the project. The CIC Metadata Portal aggregates metadata describing resources held at participating CIC institutions. Most of these objects are cultural heritage resources. Construction of the CIC Metadata Portal allowed for substantial interaction between the service provider and data providers, including exchange of shareable metadata and mapping best practices, and feedback was given on a repository-by-repository basis. Table 1 provides information regarding the size of the IMLS DCC and CIC Metadata portals. IMLS DCC Portal CIC Metadata Portal Number of records 300,000 630,441 Number of contributing repositories 35 28 Range of records harvested from contributing repositoies 35-65,000 13-300,000 Average number of records harvested per repository 7,425 25,000 Median number of records harvested by repository 1,281 6,973 Table 1: IMLS DCC and CIC Metadata Portals Dublin Core Metadata Page 3 of 24This study analyzed metadata records in the IMLS DCC portal in depth, and observations from the CIC Metadata Portal confirmed IMLS DCC findings. All records in this study were created between January 1, 2001 and December 31, 2006, and were stored and accessed on a Microsoft SQL Server. SQL queries were used for the quantitative analysis, and the qualitative analysis was performed by examining individual xml files as originally harvested. DC and OAI-PMH The decision by the OAI-PMH technical committee to require Dublin Core was controversial when first made, and continues to be seen as a negative in some settings (Cole & Foulonneau, 2007), (Lagoze, 2004), (Van de Sompel, Young & Hickey, 2003), (Chavez et al., 2006). Many in the library community are concerned about its lack of richness and specificity (Lagoze, 2001). However, one of the strengths of the schema is its ability to act as a lowest common denominator among various richer schemas, and findings indicate that use of the schema is increasing in IMLS National Leadership Grant (NLG) digitization projects (Palmer, Zavalina & Mustafoff, 2007). The Dublin Core Metadata Element Set (DCMES) has fifteen elements, all of which are optional and repeatable. These elements are Contributor, Coverage, Creator, Date, Description, Format, Identifier, Language, Publisher, Relation, Rights, Source, Subject, Title, and Type. A previous study (Shreeves et al., 2005) and best practices published by an IMLS DCC collection identify eight of these elements as