The problem of version identification in institutional repositories is multifaceted and growing. It affects most types of digital object now being deposited, and will continue to grow if left unaddressed as the proliferation of repositories continues and as they are populated with more and more content.
The JISC (Joint Information Systems Committee) has consequently funded the VIF Project as part of the Repositories and Preservation Programme, running from June 2007 to May 2008. The project is tasked with developing a framework for the identification of versions of digital objects in repositories and working to secure community acceptance of the recommendations made within the framework.
This article describes the current situation, informed by the results of two surveys undertaken as part of the project, and previews the solutions likely to be proposed by the project in the form of the framework.
The project team, through study of previous work on the issues surrounding versioning and through the use of the results of the VIF requirements exercise, identified the major issues that repository users and managers have encountered with identifying versions with repositories.
Most significant in our examination of previous work in this area were the VERSIONS Project (Versions of Eprints – a user Requirements Study and Investigation Of the Need for Standards) , and the RIVER (Scoping Study on Repository Version Identification) study  which both examined aspects of version identification in the light of the open access movement. VERSIONS, led by LSE (London School of Economics and Political Science), recently found that 59% of researchers produce four or more types of research output from each research project . Types discussed were articles, book chapters, working papers, conference papers, and presentations, amongst others. The VERSIONS project focused upon e-prints in Economics, and found that not only do researchers output these different types of object, but also that each one of these may be developed through several draft versions. These different outputs are increasingly likely to be made available as working papers, work in progress or pre-prints during the development of a piece of research, and VERSIONS found that this was leading to confusion and time-consuming inspection of objects for end-users who were trying to identify the work they were finding online.
Prior to the VERSIONS Project, the RIVER study concluded that:
'The issue of version identification is not simply (indeed not primarily) one of unique identification of resources but rather of defining the relationship between resources. While it is important that each of those resources should be uniquely referenceable, from a user standpoint the more significant questions to ask are
- In what way are these two things the same?
- In what way are these two things different (and to what extent does that difference matter to me)?' 
The number of potential versions created as part of a contemporary research project is clearly large, and the relationships between iterations, variations and manifestations are theoretically enormous. This is further complicated by projects with co-authorship and collaborative work.
Although the VIF project team has remained wary of wading into the controversial subject of how open access affects the publishing industry, it is clear from both VERSIONS  and VIF's own research that a great deal of versioning problems arise from reader confusion over the publication status of the content that they find in a repository.
The significant problem for repository managers is how best to organise these multiple versions within a repository, and how best to describe them so they can be properly found and understood. This problem extends beyond the confines of the single repository implementations and interface to include cross-repository searches and other open access harvesters. Problems identified include:
The VIF Project has moved on in two significant ways from this earlier research. Firstly, we have looked at the requirements for a variety of digital objects, not just text documents, and across the whole range of disciplines. Secondly, the VIF team has described our approach to versions as agnostic. The research carried out so far prior to VIF has taken a different approach, with much discussion already having taken place on defining and using labels to describe work in the context of the publication process.
VIF therefore defines a version as 'a digital object (in whatever format) that exists in time and place and has a context within a larger body of work'. This definition allows for ultimately user-defined opinions on what constitutes a version, and underpins the project team's desire to remain neutral and produce a clear and transparent framework. A central aim of this framework is to make available to end-users and repository managers the important information which facilitates clear and effective version identification.
The project undertook two surveys in the autumn of 2007 of over 100 information professionals working in repositories and 70 academics (mainly in the UK with some international respondents).
There are clear results from the surveys, identifying a common feeling amongst repository staff in respect of the following:
The survey highlighted some striking realities about version identification. Important to the success of the framework is the knowledge that approximately a third of information professionals with responsibilities toward a repository stated that either they have no system currently in place or 'don't know' how they deal with versioning at present.
There were few instances where opinion was shared between academics and information professionals, but both groups agreed that identification of versions is a problem – only 5% of academics and 6.5% of information professionals surveyed found it easy to identify versions of digital objects within institutional repositories. The situation becomes even worse across multiple repositories (1.8% and 1.1% respectively).
When asked what types of material they currently stored in their repositories, 95.4% of information professionals claimed that they currently store, or plan to store, text documents with many also stating that they store, or plan to store, audio files (73.6%), datasets (77.9%), images (83.3%), learning objects (46.5%) and video files (75.3%).
This awareness of a wider range of object types being created and the desire on the part of information professionals to store them are positive signs, especially in the context of the results of the survey of academics, which suggested a large number of researchers either already create or intend to create audio files (47.2%), datasets (68%), images (72.5%), learning objects (74.6%) and video files (57.6%). As expected, the vast majority also intend to continue working with text documents.
Most notable from the survey of academics was the concern that only the most complete version of their work be made available in repositories. This is in direct contrast to the wishes of information professionals, who overwhelmingly wanted to store all versions that a researcher has made available. The academics we surveyed were very clear about their wish to only make the finished version of their output ultimately available. The strength of feeling on this issue was demonstrated by the comments we received as free text, often even in answers to questions on different subjects.
A corollary of this is the behaviour therefore expected of information professionals by academics when it comes to organising versions of their work within a repository.
The project team realised that as well as addressing practical issues about version identification we will also need to examine issues such as whether repository managers should replace objects when newer versions become available, or retain all versions as representative of the research process. There are subsequent questions about the value of keeping all versions for future research purposes and preservation, and developing policies that suit individual researchers, disciplines, institutions and their repositories.
The project team offered several broad potential solutions to the problem of version identification to all survey respondents. These included chronological and numeric approaches, taxonomies and the use of id tags.
Although support was high for all of the suggestions (only one solution had less than 60% support across the respondents) free text comments revealed numerous and varied misgivings. No one solution stood out as having more support than any other. We encouraged free text responses with each suggested solution, and many qualifications and caveats were given.
The most contentious example is that of using a taxonomy to define the version status. The notion is clearly attractive, and has been addressed previously with NISO / ALPSP (National Information Standards Organisation / Association of Learned and Professional Society Publishers) , RIVER and VERSIONS all offering possible vocabularies to describe versions. Many free text comments remarked that whilst the idea is a sound one in principle, implementing such a taxonomy would be virtually impossible without some sort of enforcing body. Also, getting community agreement on the terminology used would be a very controversial subject due to the frequently polarised standpoints of publishers and information professionals. Serving the best interests of the research itself by setting up a descriptive taxonomy sounds a worthy concept, but insulating it from the pre-established terminology and bias of certain camps would clearly be a very serious undertaking.
This was a blessing and curse for the project team when developing solutions for the framework, as none of the suggested solutions had been eliminated and the options remaining were complex and by no means generically applicable. Therefore, we chose to detail many solutions, making both their benefits and problems explicit, and to allow the eventual audience of the framework to choose the solutions that suit their needs best. Ultimately the lack of a 'silver bullet' to solve the version identification problem came as no surprise when one considers the different needs across disciplines and the different types of object.
The project identified three major stakeholder audiences to address with the framework. Each has a role to play in facilitating easier version identification in the future.
The most ideal and reliable source of version information about an object must be the author or creator of that object. Awareness of versioning issues amongst content creators when disseminating their work in repositories is low, and the framework should promote better practice in both embedding version information within an object, and supplying repository staff with the right information when their work is deposited.
Much care must be taken here to keep a light touch and minimise the bureaucracy surrounding repository deposit, while still ensuring that sufficient attention is paid to identifying an object when it is disseminated in this way. The project is committed to improving the experience of using a repository, not placing obstacles in the path of users' success. It is essential therefore for VIF to consider the role of repositories through the eyes of the content creators – what versions are visible, etc.
The staff working on actual and future repositories are the key audience for VIF. Repositories across the UK and worldwide are at early stages of development, and each implementation is different. Different in terms of content stored, with different subjects, object type, collection policies. The staff time available to repositories also varies enormously, from some institutions having full time staff to others subsisting on only a few hours here and there. Therefore the framework must be implementable by staff with a range of resources available to them.
The most powerful group the VIF will address are repository software developers, as this group is able to control the exact way versions are organised and represented in repositories (acknowledging of course that much repository software is often open source and customisable by users with the right know-how).
The repository software packages currently available deal with versions in different ways. There are clear pros and cons associated with different systems. For example, Fedora allows the depositor to place new versions on top of old ones, which gives a clear identification of the newest work. However it deals less well with lateral versions, such a Microsoft Word file and a PDF file kept alongside each other. Eprints has a different approach of allowing versions to be kept in separate records that are then linked to each other. However, this only allows for a linear arrangement, as each document can only be either a successor or a predecessor of another.
A further issue concerning the repository software currently available is that even with good intentions, most are still primarily designed, or implemented, to deal with text documents. Work is going on to improve this situation and investigate ways of broadening the use of repositories to deal with different file types better, such as the DataShare Project , and the Application Profiles for Images , Geospatial data  and Time-based media. These variations will be taken into account by the VIF project team as we try to address future repository software development.
The VIF project team consists of Jenny Brace and Dave Puplett, LSE, Paul Cave, University of Leeds, and Catherine Jones, Science and Technology Facilities Council. This team has worked in conjunction with an invited expert group, details of which can be found on the VIF Web page , to develop the framework itself and the solutions that it will contain.
Having identified the three major stakeholders that the project wishes to address, the project team decided to target the framework towards each of these audiences by creating advice and recommendations specific to each group.
The first step in the chain of establishing the version status of an item should be to ensure the item identifies itself. This is fairly straightforward for text documents as they usually contain at least title and author on a front page. This can easily be expanded to include some sort of version information. This could be specific dates, a description, a version number or so on. Such information can also be embedded into the file itself by using the properties fields that most desktop applications employ. The framework gives advice on how to embed version metadata in all sorts of digital object, from text documents written in Word or converted to PDF, data in spreadsheets, through to images, video and audio created or compressed in a variety of file formats. If a document is imbued with version information by its creator, such information becomes authoritative and available to anybody who then comes into contact with the item, including the repository manager.
It is crucial for version identification that metadata stored for items with multiple and/or ambiguous versions is rich enough to support the disambiguation of versions. The project will be making recommendations on ways of achieving this, and what information on an object or set of objects in a repository needs to be made transparent to enable the end-user to establish the version status clearly.
One of the messages we will be promoting to content creators is the need to communicate version information as well as traditional bibliographic information when an item is deposited to a repository. Ideally this will also include information about anywhere else an item might have been deposited, which will begin to address the problem of version identification across multiple repositories. This message must be carefully communicated however, as it is important to minimise the burden upon authors and content creators, which is already seen as a barrier to the success of repositories.
The staff who run repositories on the ground are undoubtedly our most significant audience and the framework is at its greatest depth here. The advice offered is practical, wide-ranging and addresses several different aspects of repository maintenance. VIF will be recommending that repositories include a statement about versioning in appropriate policy documents, and will provide guidance on what to include. We have also made approaches to the OpenDOAR  team about improving the support that their template policies offer regarding versions.
The ingest and metadata capture process of an item is the best opportunity for a repository manager to establish the version status of an item. We will detail ways of incorporating version awareness into both self-deposit systems and proxy deposit undertaken by library staff. How this information is then incorporated into the metadata for an object, what parts of this information are made transparent to the end-user, and what information survives the harvesting process of cross-repository search mechanisms represent central issues that the framework must address.
We will be looking closely at how best to use current metadata standards, especially in the light of the recently completed Scholarly Works Application Profile (SWAP) (previously known as Eprints Application Profile) . The use of a FRBRised structure in the SWAP particularly appeals to the VIF Project because of the way that metadata is tied more closely, and less ambiguously, to individual 'entities', or objects as I have referred to them here. This application profile allows metadata to describe not only the work in question (such as a particular research project) but its individual versions, right down to the level of the actual format of each version and its distinct copies.
The project team will also discuss how to make the best use of other metadata schemes to describe versions, detailing how to use available fields to describe version status, and encourage their uniform usage to enhance version identification across different repositories. We will also look at recommendations for the future, to raise awareness of version identification as an issue for future metadata development projects.
The framework will also make more minor recommendations that could benefit certain repositories, such as considering implementing uniform file naming within a repository. Another solution discussed has been the use of cover sheets. Some repositories (for example LSE [13) insert a cover sheet on the front of their text documents, which can contain detailed information about the provenance of an item, including version information. However the use of cover sheets is mainly limited to text documents.
VIF's approach to developers of repository software will take the form of specifications for future software improvements that would make version identification easier. Part of this task is to improve awareness of the issue amongst this audience that version identification is a significant problem.
Some of the specifications have already been implemented by some repository software developers, but the project recommends that the following features are highly desirable across repositories.
The specifications are grouped into two broad categories; those which affect the 'on-site' usage of a repository, and those which relate to how metadata is structured within the repository ready to be searched or harvested by a third-party aggregator.
The first proposed software improvement will be to introduce automatic checking of title and author upon deposit to guard against duplication, and to help make sure versions are identified as such at this early stage. This feature would prevent an object related to another already held within the repository being deposited with the relationship being flagged up by whatever versioning structure the repository employs.
Ideally this feature would also be able to compare the deposited item to the contents of other repositories and flag up any overlap or relation to other external objects. This would be particularly useful in the case of work by multiple authors working at different institutions. The challenge for software developers is to integrate this into both templates for a self-deposit process or in the back end of the software interface for repository staff.
VIF will also recommend the wider use of thumbnailing to preview objects. This is especially useful for previewing images and multimedia items, where versions are likely to include cropped versions.
Much work has already been undertaken on the development of Dublin Core application profiles that provide richer metadata options for repositories to provide for harvesting by open access aggregating services. The framework will recommend that future repository software be designed to support richer metadata export for cross- repository searches, such as these application profiles. This could possibly include exporting a thumbnail of an object as part of the metadata, for example.
The project has identified considerable room for improvement in the way that versions are ranked by third-party searches or repositories. The very strong evidence of the VIF survey makes it clear that academics want the latest version of their work to be the one that is found first by end-users. Repositories that hold more than one version of an object may well know or record which one supersedes another, but this information is not communicated well externally, leading to erratic ranking results in external searches and confusion over which is the most appropriate version to use.
Open access has made work freely available and opened up many new possibilities for scholarly communication, but managing the way this work is identified and found is essential if repositories are to flourish.
If as we all hope, repositories continue to grow, the need for better and clearer version identification will become more and more important as more content is deposited, in various stages of development, in different forms and at diverse locations. VIF aims to raise awareness of the issues associated with version identification and offer both short- and longer-term solutions to the problems identified. Repository managers can do much to improve the situation, beginning with some recommendations that are easy to implement. Other solutions that we will be offering, such as those to software developers, are intended for the longer term.
We will recommend the framework to all three stakeholder groups and encourage its advocacy by those who are working in this community to make repositories more accessible, more popular and more successful.
If you are interested in the work of the project, please contact us or visit the VIF Web page . The framework itself will be officially launched in late February 2008 and the project team will be disseminating the framework at events through the Spring and early Summer.
I would like to acknowledge the hard work of Paul Cave, Project Officer for VIF based at University of Leeds, on the VIF survey and his subsequent report, from which this article draws results and analysis.