Introduction
Open linking has become the accepted standard in the information industry today. Publishers are providing full text of their electronic journal and book content, and content aggregators are linking from their bibliographic databases to full text content both within and outside their hosted collections. In short, accessing full text electronically is expected as baseline functionality by today’s web-savvy users. Most primary publishers either have their own websites where they offer electronic versions of their journal and book content, or contract with vendors like Lippincott Williams & Wilkins or Highwire to host their society’s website. Others, like Ovid, led the way to full text linking from bibliographic databases by introducing aggregated, fully searchable full text collection databases (Journals@Ovid), then supplemented this service by linking to remote full text with OpenLinks. Then, along came the vendors who offered content-agnostic, centralized linking servers and opened the doors to bi-directional resource linking through OpenURL. New players and solutions are arriving on the scene daily. The product offerings are complex, and sweeping new claims are being circulated that require careful scrutiny regarding initial setup and ongoing maintenance. This paper attempts to put the state of linking today into perspective.

Quantity Is King
Librarians, students, and researchers all expect to access all sources of full text and to navigate freely among their subscribed-to resources. For the most part, they don’t know or care how linking works or where the full text resides. End-users often do not know the content source or the linking format. Librarians do know that if one product does not provide the quantity and quality of full text links desired, they can find one that does.

Linking Models – Dynamic vs. Data
So, how do information providers get their users to the full text of journal articles and books? There are two basic models for linking: dynamic and data.

The dynamic or live linking model is widely used and represents the basis for OpenURL linking, a standard that, in its most simplistic definition, establishes the rules for matching bibliographic information (“metadata”). Earlier dynamic linking products, like Ovid’s OpenLinks, were developed from joint linking agreements with content providers. The result of partnering with publishers was a metadata matching algorithm that sent the A&I database searcher directly to the cited article at the publisher’s web site. Coverage tables in the Ovid product further refined the matching model to insure that the searcher saw a full text link button only for the period covered by the institutional subscription. If an institution’s electronic subscription started in 1998, any citation prior to that date would not tempt the user with a full text link that resulted in a dead end. This extra refinement requires technology on the original site and is not a part of the OpenURL standard.



The linking data model used by PubMed’s LinkOut and SilverPlatter’s SilverLinker, employs a databases of links – one record for each full text article link. The database model is arguably more reliable, and leads to fewer failed matches because a link is recorded in the database only if there is a target full text article to match the bibliographic record. However, it is also a more work-intensive model, requiring participating vendors to create linking databases, send them to the recipient system, and then to process and distribute them to all users. This takes time, which means that currency becomes the tradeoff to the less reliable, live linking model.



Universal Linking Solutions
Library automation vendors have recently added centralized linking products to their product offerings. These solutions, based on the emerging OpenURL standard of linking, offer the advantage of setting up and maintaining links to all resources through a single, centralized link repository. Linking from A&I databases to a wide variety of full text and other resources (book ordering through Amazon.com, genome databases, etc)
can be set up through a single, centralized source. Serials lists from the institution’s OPAC can be imported to simplify setup and maintenance.

OpenURL is based on a dynamic linking model and adds information in the URL query that identifies the source requester. That way, the central link server knows the origin of the link request and is able to parse it according to the rules for that vendor’s product.

However, some of these products carry a hefty price tag, and require Perl programming to build the filters to get around the syntax required by database and target resource formats. And all require that the link matching syntax keeps up with the ever-changing world of resource movement across the great digital divide.

CrossRef
The CrossRef organization began in 2000, as a collaboration of publishers with a mission to provide a reference linking “backbone” for electronic resources to promote scholarly research. Three years later, CrossRef has over 180 primary publisher members. In addition, CrossRef is used by a growing number of library ‘affiliates’ who pay for the privilege of accessing CrossRef’s metadata database to retrieve Digital Object Identifiers (DOI’s) that resolve to the article’s full text. CrossRef represents the belief that the value of a publication is tied to the number of links to and from it, and that ease of navigating from one article to another is key. That’s why publishers pay to deposit article metadata to the CrossRef database. To be a player in the electronic market is to be linked.

However, until recently, there were no methods for policing the quality of data deposited by the publishers, and there is no guarantee that a publisher has deposited all of its archival data, or is completely up to date in depositing DOI’s for current journal content. CrossRef has a worthy mission, but it’s still being proven as a long-term solution and not recommended for dynamic linking solutions. In other words, if a researcher expects that a bibliographic citation will link successfully to a full text journal article based on ISSN and publication year, then the assumption must be that the publisher has deposited DOI’s for each article in that volume and issue. When gaps exist, dead ends are the result.

Linking is Not a Perfect Science
Examples of problematic links can be found even in the best of circumstances.

The difficulty can come from any of the participants in linking -- indexers of bibliographic data, the publishers, or the service providers.

Supplements and parts can be represented and organized differently when published online; and syntax varies from publisher to publisher, and from service provider to service provider. Supplements may be indexed as "part", "pt", "supp", or "s" in volume, issue or page information.

Indexing by content providers is not standardized, so that each database uses a different set of criteria for indexing bibliographic information. Just as a publisher may organize supplements and parts differently, so do indexers. They can use any of the terms list above, or none at all. For example, MEDLINE may index a journal issue as issue “4 part 1,” while the journal issue number in the target link syntax may be “4”. A link will not work in this example because there is not an exact match on the issue information. However, the link may work correctly in another bibliographic database where the issue number for that journal is indexed as “4”. Similar problems exist with articles contained in combined issues, special issues, and supplements, all of which are indexed differently in various bibliographic databases.

It is possible to develop fuzzy matching logic, or rules to fetch additional data (e.g. article title) when the IVIP (issn, volume, issue and page) information cannot provide an exact match. The tradeoff is performance; the more processing required, the slower the link generation

The changeover from print ISSN to electronic ISSN as the journal identifier has created conflicts, as there is no standard use across information providers; where there is no standard, the quality of linking is affected. High quality linking solutions require significant resource expenditure and ongoing attention to changes. The DOI has gone a long way to establish the value of a persistent identifier in the journal linking world, but anyone who maintains an institutional link repository can attest, it’s not a one-time investment. The work is ongoing, so the tools need to be robust and flexible.

Linking to Related Information
Linking to full text does present a challenge, however it can be modeled simply in theoretical terms as the linking relationship is presumed to be one-to-one. A successful link to full text should, for the given metadata, always present a single, accessible document, initially described by originating metadata.

It is important to mention the ability to resolve link queries to the institution’s subscribed content or appropriate copy. While CrossRef provides a persistent identifier in the form of the DOI, the associated URL takes the user to a web site identified by the publisher or agent. To allow an institution to link to its locally held full text collections requires that the DOI be redirected to the local site. This capability, called “reverse metadata lookup” is starting to be advertised by vendors who offer universal link resolver products.

In today’s world, because most of the interaction between the end user and information is direct, without the assistance of an information specialist, it is extremely useful to be able to identify other documents which can be, in one or many ways, related to the document that the end user is viewing. Article metadata might contain sophisticated indexing that allows us to guide the researcher to related and probably relevant information in one or more documents in other document repositories.

To illustrate, consider the bibliographic record from Biological Abstracts database:

Title: Analysis of chromosomal aberrations involving chromosome 1q31fwdarwq53 in a DMBA-induced rat fibrosarcoma cell line: Amplification and overexpression of Jak2.
Author, Editor, Inventor: Sjoling-A {a}; Lindholm-H; Samuelson-E; Yamasaki-Y; Watanabe-T-K; Tanigami-A; Levan-G
Author Address: {a} Department of Cell and Molecular Biology-Genetics, Goteborg University, SE-40530, Box 462, Goteborg; E-Mail: Asa.Sjoling@gen.gu.se, Sweden
Source: Cytogenetics-and-Cell-Genetics. [print] 2001(2002); 95 (3-4): 202-209.
Journal URL: http://www.karger.com/journals/ccg/ccg_jh.htm
Publication Year: 2001
Document Type: Article-
ISSN (International Standard Serial Number): 0301-0171
Language: English
Abstract: In a study of DMBA-induced rat fibrosarcomas we repeatedly found deletions and/or amplifications in the long arm of rat chromosome 1 (RNO1). Comparative genome hybridization showed that there was amplification involving RNO1q31fwdarwq53 in one of the DMBA-induced rat fibrosarcoma tumors (LB31) and a cell culture derived from it. To identify the amplified genes we physically mapped rat genes implicated in cancer and analyzed them for signs of amplification. The genes were selected based on their locations in comparative maps between rat and man. The rat proto-oncogenes Ccnd1, Fgf4, and Fgf3 (HSA11q13.3), were mapped to RNO1q43 by fluorescence in situ hybridization (FISH). The Ems1 gene was mapped by radiation hybrid (RH) mapping to the same rat chromosome region and shown to be situated centromeric to Ccnd1 and Fgf4. In addition, the proto-oncogenes Hras (HSA11p15.5) and Igf1r (HSA15q25fwdarwq26) were mapped to RNO1q43 and RNO1q32 by FISH and Omp (HSA11q13.5) was assigned to RNO1q34. PCR probes for the above genes together with PCR probes for the previously mapped rat genes Bax (RNO1q31) and Jak2 (RNO1q51fwdarwq53) were analyzed for signs of amplification by Southern blot hybridization. Low copy number increases of the Omp and Jak2 genes were detected in the LB31 cell culture. Dual color FISH analysis of tumor cells confirmed that chromosome regions containing Omp and Jak2 were amplified and were situated in long marker chromosomes showing an aberrant banding pattern. The configuration of the signals in the marker chromosomes suggested that they had arisen by a break-fusion-bridge (BFB) mechanism.
Abstract Indicator: Y
Major Concepts: Methods-and-Techniques; Molecular-Genetics (Biochemistry-and-Molecular-Biophysics); Tumor-Biology
Super Taxa: Muridae-: Rodentia-, Mammalia-, Vertebrata-, Chordata-, Animalia-
Organisms: LB31-cell-line (Muridae-): rat-fibrosarcoma-cells
Taxa Notes: Animals-; Chordates-; Mammals-; Nonhuman-Mammals; Nonhuman-Vertebrates; Rodents-; Vertebrates-
Parts, Structures and Systems of Organisms: rat-chromosome-1: q31-q53-region
Chemicals and Biochemicals: 7-12-dimethylbenz[a]anthracene-: carcinogen-; PCR-probes [polymerase-chain-reaction-probes]: probe-
Sequence Data: AF054619-: GenBank-, nucleotide-sequence; D14014-: GenBank-, nucleotide-sequence; L29232-: GenBank-, nucleotide-sequence; M13011-: GenBank-, nucleotide-sequence; M26926-: GenBank-, nucleotide-sequence; S78355-: GenBank-, nucleotide-sequence; U03184-: GenBank-, nucleotide-sequence; U13396-: GenBank-, nucleotide-sequence; U49729-: GenBank-, nucleotide-sequence; X14849-: GenBank-, nucleotide-sequence; Y00848-: GenBank-, nucleotide-sequence
Diseases: fibrosarcoma-: chemically-induced, genetics-, neoplastic-disease
CAS Registry Number (R): 391544-71-1: GENBANK-U03184; 160102-94-3: GENBANK-U13396
Methods and Equipment: ABI-377-DNA-Sequencer: ABI-, medical-equipment; BigDye-cycle-sequencing-reaction: PE-Biosystems, medical-equipment; PCR- [polymerase-chain-reaction]: DNA-amplification, DNA-amplification-method, analytical-method, genetic-method; Southern-blot-hybridization [Southern-blot]: analytical-method, detection-method, gene-mapping, genetic-method, labeling-, recombinant-DNA-technology; comparative-genome-hybridization: Molecular-Biology-Techniques-and-Chemical-Characterization, analytical-method, genetic-method; dual-color-fluorescence-in-situ-hybridization-analysis [dual-color-FISH-analysis]: Molecular-Biology-Techniques-and-Chemical-Characterization, analytical-method, gene-mapping-method; fluorescence-in-situ-hybridization [FISH-]: Molecular-Biology-Techniques-and-Chemical-Characterization, analytical-method, gene-mapping-method; radiation-hybrid-mapping: Molecular-Biology-Techniques-and-Chemical-Characterization, analytical-method, gene-mapping-method
Miscellaneous Descriptors: break-fusion-bridge-mechanism; chromosomal-alterations; comparative-gene-maps; marker-chromosomes
Alternate Indexing: Fibrosarcoma-(MeSH)
Accession Number: 200200260598
Update Code: 20020909

In the above record we can use the information in the Organisms field to link to more detailed taxonomy information in the taxonomy database, Sequence Data field content to display more information on genes mentioned in the article from the Genome databank, CAS registry number for more information on physical and chemical properties of substances mentioned in the article, content of the field Methods and Equipment can be used to link to related information in the MethodsFinder database, Diseases field can be used to automatically link to clinical management of the mentioned disease, while Alternative Indexing field can be used to link to related information in any other database using MESH indexing.

It has to be said that there are very few systems which currently support this type of linking. Even OpenURL does not yet provide sufficient granularity of the metadata to standardize this type of linking.

Linking Futures
Bibliographic database providers may eventually incorporate the DOI as a field identifier to link bibliographic database citations to full text at the publisher’s web site. Several information providers are doing this now and others are sure to follow. Then, the mandate will be to ensure that there are solutions that account for appropriate copy management, or redirecting the DOI identifier to an alternate URL that points to an institution’s locally held or aggregator-supplied content.

Conclusion
Using the linking methods outlined above, most A&I vendors can offer certain basic linking functions from their own bibliographic content, and the expectation is that they will be available at no extra cost. For those entertaining the prospect of centralizing link administration, here are some things to look for:

An optimal centralized link solution should:

• Support the OpenURL standard for bi-directional linking that includes linking to any full text, document delivery, catalog holding and other OpenURL compliant internet resource.
• Address linking to the appropriate copy of a resource, redirecting CrossRef and DOI links to locally held resources by means of maintaining local resolution tables.
• Be priced competitively. Compare before you buy. There are a number of universal linking products on the market. Ease of administrative setup and maintenance are important.