Persistent identi cation is necessary for recognition, dissemination and (external) cross-references to digital objects. Uniform Resource Identi ers (URIs) provide an established scheme for this task, but do not guarantee stable and persistent identi cation. In the context of (personal) archives, stability is needed when references are be stored on a medium where later changes to identi ers cannot be corrected at all or only with a very large overhead, such as WORM media or tape archives. Additionally, resources like contacts or appointments do not have a URI, while other URIs, such as le system paths or the IMAP URI, are unstable by design and cannot represent the dynamic aspects of Personal Information Management (PIM). This paper discusses problems of archiving that arise with entity identi cation in PIM, especially on the example of the personal le system.
During the past decades, our everyday work-life has rapidly moved from the real world to the digital domain. Then we used to put our hands on paper les, books, business cards, calendars or photographs, nowadays everything is a digital object stored on the computer, in the smart phone or the Cloud. Still we need ways for virtually grabbing and referencing those objects.
The Uniform Resource Identi er (URI) serves exactly this need: to identify digital resources for later look-up and to reference objects. However, there is only a very generic speci cation for how a URI must look like. This results in high exibility and viability for many applications, but URIs may not match the requirements posed by the task. For example, the proposed URI for a le path is unstable in the way that those URIs become invalid whenever the le is moved. Therefore external references in objects no longer point at that le and have become stale and thus useless.
Even worse, if these objects are archived, changes to the stored data are
often prohibited either by policy or due to the storage medium. For example, A
CD-ROM is read-only and re-writing a tape storage takes much e ort. Therefore
identi ers need to be stable so that existing archives are not broken.
Having reliable identi ers for digital objects is a requirement to mine and
store relationships between those entities, especially in dynamic environments
such as Personal Information Management. With an integrated view novel
interaction concepts, such as graph based interaction, are possible. These approaches
enable the user to explore large graphs{such as the graph emerging from
connected objects in Personal Information Management{in order to nd speci c
elements without the need for keyword-based search or to get an overview on
structures in the personal information space (e.g. [
The paper starts with a short analysis of related work and a de nition followed by an elaboration of requirements and problems with stable identi cation systems. The ndings are further discussed on the example of a personal le system to point out some problems that arise with current systems. 2
Entity identi cation has uses in several areas: In geometry, entity identi cation
is a known problem when objects change, but reference points must be
recognized [
No related work could be found towards building stable identi ers in Personal Information Management and regarding personal archives, hence this paper aims at starting a discussion in this direction by identifying some general problems based on examples of stable URIs and personal le systems. 3 3.1
A general de nition for the term identi er can be found in [
Any association of a name with a thing { by anyone { establishes an identi er. A name is not an identi er unless it identi es something. (E.g. an unassigned phone number is a name, but not an identi er.) Digital identi ers exist in contexts which are seldom made explicit or included when citing an identi er.
RFC 3986 [
When looking for an identi er, the URI makes a good choice. Sharing the syntax with URLs, any entity|real, abstract or on the Worldwide Web|can be identi ed without introducing a new standard. 3.2
The URI speci cation de nes requirements regarding syntax and semantics of a
URI. However, it is agnostic regarding requirements for a system using the URI.
This leads to high exibility, but demands further analysis in the context of an
application using the URI. In [
Global uniqueness Identi ers are unique, i.e. one identi er does not reference information associated with multiple resources.
Persistence Identi ers uniquely reference resources beyond their lifetime. This speci cally means that an identi er must not decay when the resource is no longer available and, in a broader interpretation, that an identi er must not be re-used for a resource once it has been used.
Scalability Identi ers are scalable and can be assigned to any resource. Legacy support Identi ers can support legacy identi cation schemes to the extent that these satisfy minimum requirements.
Extensibility Identi cation schemes can accommodate future extensions. Independence Responsible authorities maintain and assign resource identi ers within a given system.
Resolution Identi ers are supported by services that enable their translation.
The W3C demands opacity for all identi ers [
Axiom: Opacity of URIs The only thing you can use an identi er for is to refer to an object. When you are not dereferencing, you should not look at the contents of the URI string to gain other information. URI opacity has been a source of many debates, especially since technical solutions like the GET method for submitting HTML form content explicitly violates this rule.
A more ontological approach towards identi er requirements can be found
in [
In order to obtain the resource denoted by an identi er, the identi er must be
resolved to a locator which allows to retrieve the resource [
First, identi er schemes based on Uniform Resource Names (URN) as
dened in RFC 2141 [
A second option are handle-based systems: Instead of encoding the
information needed to nd an entity into the identi er, a handle to a database entry
in a resolver is created. In order to retrieve the entity, i.e. resolve the identi er,
a look-up call to the resolver is necessary. The identi er is then transformed in
either the resource itself or a di erent identi er which can be used to retrieve the
entity. Among many, a well-known handle-based example is the DOI R System,
maintained by the International Digital Object Identi er Foundation (IDF)2.
The IDF hands out unique pre xed for publishers, who create unique identi ers
in their own namespace. Thus each digital document, such as publications in
conference proceedings, can be referenced without knowing its actual storage
location. The DOI resolver maps a certain DOI to a URI in the database, which
can be used to retrieve the document. In this case, the resolution consists of
multiple steps, as the IDF does not store the documents themselves but identi er
for the storage locations. Other known handle systems are Persistent Uniform
Location Locators (PURL) 3 and the Archival Resource Key4. Please refer to [
Many systems are handle based and when the task of resource identi cation arises, handles and central registries seem to be the preferred choice. In the context of Internet connectivity and centralized systems this makes sense: Having a control entity avoids naming clashes and makes it simple to decouple identi er and resource location. However, a handle based system also requires the availability of a handle database. In the case of personal archives, it may be necessary to store the resolution database alongside with the archive. This may be possible for a complete snapshot with stable references, but if the references may still change, the archived database will be outdated and if only parts of the personal information is archived, the resolution database must either be split or very likely results in a large storage overhead. For an o ine scenario or when data is only available locally, e.g. on the personal computer, a handle
database may not be feasible at all. Here the information needed for resolution must become a part of the identi er itself. There are URI speci cations for many identi ers in the PIM context, such as le paths or e-mails, but these often do not take stability into account.
The following case study shows problems that may arise when identifying resources in the personal le system. 4
Hierarchical le systems are still the main way of storing information on a personal computer. PIM systems have to provide means of identifying those les in order to create stable references.
Several systems have been devised towards le identi cations, all with their own strengths and caveats. A small selection shall be discussed to give an overview on the problem eld.
File Path A naive approach is given by the le path, i.e. the location of the
le in the local le system. RFC 1738 [
Magnet Links A widespread solution for referencing les based on their content
is provided by magnet links6. Although not listed as a standard, the magnet:
pre x can be found in the list of URI schemes7. Instead of a le path, the URIs
are generated from a hash code computed from the binary content of the le,
such as SHA-1 [
Unix -based systems. (http://www.pathname.com/fhs/pub/fhs-2.3.html) 6 http://magnet-uri.sourceforge.net/ 7 http://www.iana.org/assignments/uri-schemes/uri-schemes.xhtml URIs are not very common on local le systems for two reasons: First, they have a rather negative reputation, as these identi ers are mostly used in P2P le sharing. Here they serve the purpose very well, as they identify a certain content without knowledge of its location. Second, magnet links become invalid if the le content changes{another intended use of les in Personal Information Management.
Heuristics When both le path and le content change, heuristics must be used
to identify a le. The GIT version control system 8 tracks changes in the le path
by comparing the content of edited les, i.e. if a le is removed and a new le with
su ciently similar content appears, the change is logged as \renamed" instead
of \new". Similar, changes in the le path could be detected by monitoring
the le system operations, which is supported by common operating systems.
Methods from duplicate detection can be employed to restore links by providing
candidates to replace a broken le reference. Spinellis [
Version Control Systems In an assumed system where content cannot be deleted but only updated, such as found in a version control system, the problem boils down to two distinct cases: 1. The generated identi er references a certain version of a le, i.e. a stable content. 2. The generated identi er references a certain path of a le, i.e. move or rename operations will not be applied.
User ,ce rse fa ow re rB t In le r i se .F U .ge
file: magnet: broken link
Hash Index Lookâup heuristic
File System To create a stable identi er the intended use of a le must be known to select the right approach. Figure 1 shows how the described components could work together to resolve le links. In current operating system this leads to a semantic gap between the available data and the user's intention. It is left to future research to determine if and how this gap can be closed, e.g. based on the le type, by observation of user interaction with the le or by taking the le's provenance into account. 5
In this paper we have discussed some fundamental problems that arise when persistent identi ers for objects from the PIM domain are needed. Requirements and modeling of identi ers cover a growing research eld, however, PIM solutions still rely on locally generated identi ers instead of stable, global URIs. In a case study some problems with persistent identi ers have been shown: Stable le links require more attention and may even lead to a semantic gap where the system cannot decide how to generate the correct link.
Future work includes research on remaining PIM entities, speci cally e-mails, contacts and appointments as well as further tests on implementation of these identi ers and tests towards reliability in di erent data sets. The le case study has shown that reliable persistent identi cation will need methods from machine learning to repair broken links and determine the intended use of an object. Semantic archives may provide the necessary information for re-assigning an object to its previous link if a connection has been lost, e.g. by comparing metainformation. Finally, when persistent identi cation is possible, user interfaces and semantic desktop applications can make use of those identi ers.
Parts of this paper have been researched within the scope of the SENSE project which is funded by the Federal Ministry of Education and Research, German Aerospace Center. It is part of the KMU-Innovativ: IKT campaign and goes by the funding numbers FKZ 01IS11025E.