Metadata Change in Traditional Library Collections and Digital Repositories: Exploratory Comparative Analysis

Transcription

1 Metadata Change in Traditional Library Collections and Digital Repositories: Exploratory Comparative Analysis Oksana L. Zavalina Priya Kizhakkethil Shadi Shakeri ABSTRACT This pa reports preliminary results of a comparative study of metadata change over time in two different environments digital repository and bibliographic utility used for cooative cataloging -- and with two different metadata schemes a local versio Dublin Core and Machine Readable Cataloging (MARC). The findings with regards to metadata change types and subtypes observed, as well as the most frequently occurring change categories and metadata in which the change occurs most frequently are presented, and the implications are discussed. Keywords Metadata aggregations, metadata evaluation, metadata change, digital libraries, digital repositories, MARC, Dublin Core INTRODUCTION AND BACKGROUND To ensure that the users successfully find, identify, select and obtain information objects that they need, high-quality metadata should be provided to adequately represent information objects contained in both physical and digital libraries. In the digital library research community, a variety of metadata quality criteria have been suggested to guide metadata management and evaluation, with the three most widely accepted criteria identified as accuracy, consistency, and completeness (Park, 2009; Park & Tosaka, 2010). Metadata accuracy is measured as the degree to which the data values in metadata match characteristics of the described information object (e.g., ASIST 2015,November 6-10, 2015, St. Louis, MO, USA. Authors retain copyright. Stvilia et al., 2007). To fully assess metadata accuracy one needs to examine the information object itself and compare it with the metadata representing it which is a very time-consuming process; therefore most of the metadata quality studies assessing metadata accuracy look at easilyidentifiable and quantifiable aspects of accuracy such as presence (or lack) of typographical errors, etc. Metadata consistency is further subdivided into semantic and structural consistency (Park, 2009). Semantic consistency refers to an extent to which the same values or elements are used for representing similar concepts. Structural consistency is evaluated as a degree to which the same structure is followed in representing information in certain metadata elements (Bruce & Hillmann, 2004). Metadata completeness the third most important metadata quality criterion (Park, 2009) -- is evaluated as an extent to which objects are described using all applicable metadata elements to their full access capacity. Some of the assessment criteria used to evaluate metadata completeness (Moen, Stewart & McClure, 1998) include the number of metadata elements, practice of presenting blank metadata, use of mandatory and optional elements in metadata s. Many of the metadata quality criteria and measures to evaluate them which were identified in digital library community are equally applicable in digital library metadata (created in Dublin Core, Metadata Object Description Schema / MODS, Visual Resources Association Core, and other digital library metadata schemes) and traditional library metadata (catalog s created in Machine Readable Cataloging / MARC bibliographic metadata scheme). However, in the traditional library metadata community, little empirical research has been conducted to date that analyzed the quality of library metadata. In particular, the quality of library metadata s created according to the international cataloging code that replaced the previous standard, Anglo-American Cataloging Rules, in 2013 Resource Description and Access (RDA) has not been investigated. The only

2 available exploratory study of the quality of RDA MARC library metadata to date (Zavalina, Miksa, Shakeri, & Brown, n.d.) adopted the top-level computational approach prevalent in digital library metadata quality studies. According to Ochoa and Duval (2009), most of the digital library metadata quality studies involve content analysis on statistically significant samples of metadata s. Collection-level metadata s that describe entire collections of information objects as a whole as opposed to individual objects can still be examined manually due to the reasonable numbers of metadata s to work with. However, with the rapid growth of digital libraries and repositories that aggregate hundreds of thousands and often millions of items and their respective item-level metadata s, the evaluatio much more numerous item-level metadata when it is done relies mostly on computational approaches. This results in the prevalence of top-level quantitative analysis and the lack of more in-depth qualitative analysis of item-level metadata. Some of the metadata research suggests that metadata change can be viewed as indicatio metadata quality and as such is worth examining to help improve the quality of metadata. For example, as part of a study of collection-level metadata quality in the IMLS DCC aggregatio IMLSfunded digital collections, a group of researchers conducted longitudinal analysis of the modifications that had been made by digital collection develos housed at various cultural heritage institutions throughout the United States to collection-level metadata s created by hosting institutions staff in the IMLS DCC (Zavalina et al., 2008). They found that the data values associated with the Dublin Core Collections Application Profile s Subject, Audience, Size, Spatial Coverage and Temporal Coverage metadata elements are modified the most frequently, thus indicating that subject metadata (including topical, geographical, and temporal) is in the focus of metadata quality improvement efforts. Another, recent, study (Tarver et al., 2014) was the first study to focus on examinatio metadata change in a digital library. The high-level computational analysis of metadata change in the Portal to Texas History conducted as part of Tarver et al. (2014) study, complemented by Zavalina & Kizhakkethil s (2015) in-depth content analysis study allowed to develop the draft general model of metadata change in digital libraries (Zavalina et al., 2015). The exponential growth of library collections, supported by federal, state and local funding requires high-quality metadata to make materials collected by libraries fully accessible. However, investigation into the library metadata quality has not yet kept up with the demand for it. In particular, no research studies to date conducted in-depth content analysis of metadata change in traditional library catalogs and attempted to build a model of library metadata change. The study preliminary results of which are reported here relies on grounded theory approach (Glaser & Strauss, 1967) to address the research gap identified above. METHODS The research method used in this study is a manual content analysis of traditional library metadata s created in MARC metadata scheme according to the new RDA cataloging rules (specifically the bibliographic s that have undergone changes) obtained from the WorldCat database ( and comparison with the results of the similar analysis of DC-based metadata in a digital repository obtained in the related study (Zavalina & Kizhakkethil, 2015; Zavalina et al., 2015). The following research question guides this investigation: What are the characteristics of RDA-based MARC library metadata change over time and how similar or different are those from metadata change in Dublin Core (DC) digital repository metadata? In particular, the study aimed to identify in both environments metadata and sub that are changed the most often and how they change. The study also measured relative frequency of occurrence of major categories of library metadata change such as Addition, Deletion, and Modification and their respective subcategories. The sampling of DC digital repository metadata s and RDA-based MARC library metadata s was based on the number of revisions of a library metadata : the s that have underwent at least one editing event were selected. Digital repository DC s analyzed in this study were created since late 2009 and late 2012 as the versioning capability of the digital repository system which allows capturing all versions of a metadata was introduced in the fall of The RDA-based MARC library metadata s were selected for this study as opposed to AACR2-based MARC library metadata s for two reasons: the currency of the standard (all AACR2- based s will eventually be transformed into s complying to RDA cataloging rules, and efforts on both manual and automatic transformatio millions of AACR2-based s are already on the way in the library community) and relatively recent time of creation which is comparable with dates of creatio DC s in this study (fist RDA-based s started appearing in WorldCat in 2010 as part of preparation for the shift from AACR to RDA cataloging rules that occurred in 2013), and the assumed (through anecdotal evidence) high. The DC digital repository metadata was obtained from the Portal to Texas History develos in XML format, with each in a separate XML document. The RDA-based MARC s for video ings in DVD format were collected at two points in time the 2013 versio s in MARC XML format were obtained from OCLC Research team in a single file consisting of 932 RDA-based MARC s matching the sampling criteria and the 2015 versions of a randomly sampled sub-set of 369 of these same s were imported from WorldCat through OCLC Connexion tool. The RDA-based MARC library metadata s were transformed with the help of MARC Edit software into a mnemonic human-readable MARC format.

3 The earlier and later versions of each metadata were manually compared side-by-side using Sublime editor and/or other text editors. The change over time in digital repository metadata s was analyzed in 2014, and the categories of metadata change which emerged from that analysis informed the preliminary list of categories for analysis of change in MARC library metadata s over time which has been conducted in These categories were adjusted for MARC metadata and more categories were added to this list -- which was used as a coding manual in the study presented in this pa based on researchers knowledge of MARC standard and exiences working as creators and evaluators of MARC metadata. Table 1 presents the categories and subcategories of MARC library metadata, along with examples for each subcategory. Four additional subcategories of MARC metadata change which emerged in the process of analysis all belonged to the Modify broad category and are listed in the highlighted section at the end of the table. The analysis of change over time in RDA-based MARC library metadata s began with the preliminary analysis of a small separate pilot sample (15 s) with the purpose to verify and refine the draft the model of metadata change and coding manual. Next, content analysis of the main sample was formed. The preliminary results are reported in the next section. FINDINGS AND DISCUSSION Table 1 shows the categories of metadata change observed in RDA-based MARC s, with examples and codes used in analysis. Table 1. Metadata change categories and subcategories observed in RDA-based MARC library metadata Change Code Example of category category Addition: Add new ANF add a 830 (series title added access point) not previously in the Add new sub to existing ANS add a sub $e to 300 or 040 or add a sub $b to 338 Add a 2nd+ Add a 2nd+ sub Add an indicator to existing Add data value to fixed sub AEFI add a 4 th instance of 650 (topical subject heading) or a 2 nd and 3 rd instances of 655 (genre heading) AESI add a 2 nd instance of sub $a to 264 to represent additional publisher or add a 3 rd instance of sub $e to the same 700 (sonal name added access point) for a relator term to represent contributor's additional role AIF add a first indicator 1 for existing 041 (language) _ -- to indicate the item is a translation AD add a data value "g" in previously empty Audn sub to indicate general audience Deletion: Delete a completely Delete a sub completely Delete a 2nd+ Delete a 2nd+ sub Delete an indicator from existing Delete data value from fixed sub Modification: Replace an indicator in existing with new Replace tag with new for the same data value Replace existing data value with new Amend existing data value Transpose multiple instances of the same Transpose sub within a Transpose different within a Merge content with another Break an existing into separate DF remove the only instance of 250 DS DEFI remove the only instance of sub $c (statement of responsibility ) in 245 or remove the only instance of the sub $e (relator term) in 710 (corporate name added entry) delete the 4 th instance of 500 (general note) DESI delete a 3 rd instance of sub $a in the same 043 (geographic area code) DIF remove 2 nd indicator "0" in 505 to represent the fact that contents note formatting is basic not enhanced DD remove a data value in the optional 2 nd instance of Dates sub which is designed for holding original date of publication when describing reprints MRIF change the 1 st indicator in 490 from "0" to "1" to indicate that the title of the series I s established as a uniform title in the controlled vocabulary and that 830 with series title added entry is included MRFT change tag from pre-rda 260 to RDA-based 264 or change tag from no-longer-valid 440 to 490 while keeping the same data value MRD replace "Motion picture" with "Television program" in 380 or replace "a" with "i" in Desc sub of a fixed MAMD change "two-dimensional moving image" to "three-dimensional moving image" in 336 ; change "160" to "165" in Time sub of a fixed MTIF move the 700 (sonal name added access point) with the name of film director above the other 700 with names of the cast or move the instance of 264 holding copyright information above the instance of 264 holding production information MTS move the 041 $j before 041 $h MTF move 505 (contents note) from below to above 520 (summary note) MMCF merge the data value from 500 with the data value in another note (e.g., 538) and delete this instance of 500 MBFS Break down the content of a single instance of a 505 containing information on the content of three CDs into three instances of 505 each containing information on one CD

4 Break an existing sub into separate sub Replace sub indicator with new MBSS Split the data value of 300 $c which incorrectly includes both dimensions and additional material information between 300 $c for dimensions and 300$e for additional materials) MRCI Change 700 $4 to 700 $e and keep the same data value The categories of change of digital repository metadata in a local customized versio qualified DC overlapped with those for MARC metadata s. However, due to the unique nature of MARC metadata that includes both variable and fixed, the DD category (delete data value from fixed sub) was not applicable to DC digital repository metadata. The same is true about the AIF, DIF, and MRIF categories which deal with another unique feature of MARC metadata scheme (a.k.a. format) which includes numeric indicators for variable. On the other hand, several categories of metadata change were observed in Dublin Core digital repository metadata s and not observed in MARC library metadata. These included addition, deletion, or modificatio a DC qualifier or XML attribute of a. Because of the nature of MARC standard which is not using XML attributes or qualifiers, these we not applicable to MARC metadata. Other categories of metadata change not applicable to MARC library metadata but applicable to DC digital repository metadata were dealing with empty and likely specific to the digital repository. They included the modification in which the previously empty is populated, and the deletio previously empty. To the contrary, in MARC metadata, all empty are deleted upon adding the to the database, which results in the absence of empty in the, so these two categories of metadata change become not applicable to MARC metadata. Of the categories of metadata change presented in Table 2 above, the following eight subcategories (in descending order) have been observed the most often with at least 100 observations of each change subcategory in RDAbased MARC metadata s: 1. Add new sub (ANS) 2. Add new (ANF) 3. Transpose within a (MTF) 4. Add new instance of an existing sub (AESI) 5. Transpose multiple instances of the same (MTIF) 6. Transpose sub within a (MTS) 7. Amend data value (MAMD) 8. Delete (DF) 9. Delete instance (DEFI). It is easy to see that various kinds of additions clearly prevailed in MARC library metadata (they accounted for close to 50% of all metadata change instances). Modifications were the second most widely observed, with a lion share of these being transpositions of different kinds. In the digital repository DC metadata s, deletions were observed the most often, followed by modifications and additions. One of the listed above top subcategories of metadata change in MARC library metadata was also observed in DC digital repository metadata deletio (mostly those that were empty in the initial versio metadata ). Unlike with MARC metadata, transpositions of any kind were observed in DC metadata very infrequently, and additions were mostly of 2 nd or further instances of a or sub not a completely new or sub. Table 2. Metadata change distribution: variability Environ ment Portal to Texas History World Cat Range Mean Median min max max % of % % The maximum centage of in the (Table 2) was higher for DC metadata in digital repository environment. It is interesting to see that despite the fact that MARC s are naturally much longer than Dublin Core s due to the drastic differences in richness of the two metadata schemes our analysis shows that MARC s contained up to 40 non-empty while DCbased Portal to Texas History s had no more than 21 the median number of metadata were the same in MARC and DC s. The mean number of changed, however. was higher in MARC metadata. Another difference in quantitative characteristics of metadata change in the two environments is that the minimum number of with changes in RDA-based MARC library metadata s is higher than in Dublin Core digital repository metadata s. This is due to the fact that every time any change is made to MARC, 040 in MARC s is automatically updated by the software a new instance of sub $e with the code of institution that updated the is added to this. The ten most metadata in RDA-based MARC library metadata with change observed in at least 10% of metadata s -- included the following (listed in descending order): Topical subject heading Carrier type Media type Content type Genre subject heading Geographical subject heading Geographic area code Corporate name subject heading Personal name subject heading Added sonal name entry.

5 Three of these, occurring under numbers 3-5 are the new RDA dealing with content, media, and format of the item. These were not part pre-rda MARC s. Five others are subject that are used in both pre-rda and RDA s: topical subject, genre subject, geographic subject, corporate name subject, and sonal name subject. Another type of sonal name access point for the contributor (700) also was among the top 10. Additional geographic the 043 which holds the machine-readable code for the geographic location the item is about was also frequently changed. In addition, changes in 040 Cataloging source were observed in 100% of s this is automatically updated when anything is changed in the in the. Unlike in RDA-based MARC library metadata, in DC digital repository metadata s, subject Coverage and Subject underwent updates somewhat less often overall and were the 6 th and 8 th most frequently. The five most DC were Creator, Contributor, Publisher, Identifier and Note. Only one of these Contributor (700) was observed among top 10 most frequently changed in RDA-based MARC library metadata in our study. Three DC metadata elements Creator, Publisher, and Description were the most modified elements. Four metadata elements Coverage, Contributor, Primary Source, and Relation underwent the most deletions. Additions most often occurred in Identifier, Note, and Subject metadata elements. CONCLUSION The metadata community needs a common vocabulary to discuss different kinds of metadata change. This will make it easier to compare versions within a system, and to discuss and compare metadata changes across different systems. Successful digital curation strategies involve mechanisms for both pre- and post-ingest metadata normalization and quality control. Understanding and managing metadata quality requires a cyclical process that balances the evolving needs of the users, the requirements of national/international standards, and the local environments of the metadata creators. With this in mind, the ability to accurately describe and communicate change events during the lifecycle of metadata associated with a digital object will be increasingly important. The study some results of which are presented above developed an empirically-based model of metadata change which categorizes metadata change in MARC library metadata s over time. This model has been tested and is applicable to analysis of billions of existing MARC metadata s either RDA-based or pre-rda. The study reported here and future research building on its results are expected to make important contributions in building understanding of metadata change and its relation to metadata quality in library catalogs and digital repositories alike. REFERENCES Bruce, T.R., & Hillmann, D.I. (2004). The continuum of metadata quality: defining, expressing, exploiting. In Hillman, D. and Westbrook, L. (Eds.), Metadata in Practice. Chicago: American Library Association, pp Glaser, B., & Strauss, A. (1967). The discovery of grounded theory: Strategies for qualitative research. Chicago: Aldine. Moen, W.E., Stewart, E.L, & McClure, C.R. (1998). The Role of Content Analysis in Evaluating Metadata for the U.S. Government Information Locator Service (GILS): Results from an Exploratory Study. Retrieved from tanalysis.htm Ochoa, X., & Duval, E. (2009). Automatic evaluatio metadata quality in digital repositories. International Journal of Digital Libraries, 10, Park, J. (2009). Metadata quality in digital repositories: a survey of the current state of the art. Cataloging & Classification Quarterly, 47 (3), Park, J. & Tosaka, Y. (2010). Metadata quality control in digital repositories and collections: criteria, semantics, and mechanisms. Cataloging & Classification Quarterly, 48 (8), Stvilia, B. et al. (2007). A framework for information quality assessment. Journal of the American Society for Information Science and Technology, 58 (12), Tarver, H., Zavalina, O.L., Phillips, M., Alemneh, D., & Shakeri, S. (2014). How descriptive metadata changes in the UNT Libraries Collections: a case study, Proceedings of the International Conference and Workshop on Dublin Core and Metadata Applications, Austin, Texas, October 8-11, Zavalina, O.L., Kizhakkethil, P., Alemneh, D., Phillips, M., & Tarver, H.S. (2015). Building a framework of metadata change to support knowledge management. Journal of Information and Knowledge Management, 14 (1). Zavalina, O.L., & Kizhakkethil, P. (2015). Exploratio metadata change in a digital repository. Proceedings of the iconference Newport Beach, California, March 24-27, Zavalina, O.L., Miksa, S., Shakeri, S., & Brown, M. (n.d., under review). Cataloging over time: library metadata quality and metadata change of RDA-based s in OCLC. Journal of Library Metadata. Zavalina, O.L., Palmer, C.L., Jackson, A.S., & Han, M.-J. (2008). Evaluating descriptive richness in collectionlevel metadata. Journal of Library Metadata, 8 (4),