Journal reference: Computer Networks and ISDN Systems, Volume 28, issues 711, p. 1027.
This paper to be submitted to the Fifth International World Wide Web Conference.
Document last altered by Les Carr at Mon Jan 29 17:56:10 GMT 1996
This paper contrasts this approach to link management with the more generally accepted (and safer) solution of a closed information environment (perhaps in the form of an object-oriented SGML database), and presents an expanded open hypermedia service for the WWW composed of link management together with document management and consistency maintenance tools.
The Distributed Link Service [Carr95], [Carr94] provides a link management and delivery service to augment the Web's document delivery service. It is characterised by providing databases of links (linkbases) which can be applied to documents of various kinds (HTML Web documents, plain text documents, Acrobat PDF etc.). The philosophy is that links should be able to be applied to all kinds of application-specific document types, either by publication-time compilation of the links into the application's proprietary hypertext link format, or via browse-time user queries on key-words, -phrases or -sections [Hall94].
Each linkbase in the DLS makes links between source and destination documents which can in turn be stored on other sites. Although it conforms well to the HyTime [HyTime] model of independent links, it raises problems of keeping track of the documents to which each link points, as well as making sure that the link stays consistent with the (possibly changing) contents of the remote documents.
As if that is not enough, complications are added by the functionality of the DLS itself. Although the original versions of the software assumed that the linkbases would be resident on the host that was running the server software, later versions allow the server to access linkbases stored anywhere on the Web by specifying a URL. So we see a situation where not only the source and destination documents are not under the authors control, but neither are the links themselves (see Figure 1)!
|Figure 1: Relinquished Control in a Distributed Hypertext Environment|
1a: User controls all documents and links
1b: User controls few documents and all links
1c: User controls few documents and few links
Initially there is a problem simply of keeping track of the network resources that are being used and depended upon: a hypermedia information environment starts by making use of documents on the local site under the author's jurisdiction. As it develops it may start to include references to online glossaries or databases, and then to relevant online journal papers. It is obviously desirable to maintain a list of materials which are used (or pointed at) in this information environment, partly from the end-user's point of view, to help them form an impression of the scope of the material that they are looking at and providing them with an alternative navigation mechanism [Ayers95]. This also helps the author, since both producer and consumers of a hypertext need to maintain a user model of the information.
Beyond this requirement is the need to maintain the information environment, to check the availability of the remote resources, and to ensure the continued validity of the links (this situation is of course familiar to any user of the Web who maintains a list of links to their favourite Web sites). In the following sections we will discuss the use of some simple document and link management and maintenance software to help with these tasks in the context of creating hypertexts (or webs) which are designed to span many domains of authority: i.e. authoring beyond the confines of a single site.
The process of authoring is usually done with material under one's control, allowing an individual (or a co-operating group of individuals) to apply their own authorship, narrative or summative skills to create a web of information. Hypertext research literature describes a hypertext as being composed of different regions of coherence [Moulthrop92] (where the coherence comes from the authors' influence) and classifies them as as hyperbases or hyperdocuments according to the degree of deliberate authored influence in the network of nodes [Stotts91].
This poses an interesting challenge for the Web's semi-co-operative global environment: is it possible to expand the coherence found in single document or in a supervised hypertext (created by an individual or planned collaborative effort) into a some form of global coherence? Or must a hypertext necessarily fragment into a hyperbase beyond a certain scale?
Experience has shown that Web documents are frequently divided into content-bearing documents (typically not making much use of links) and catalogue documents, not containing much content, but containing a large number of links. In effect, the coherence exhibited in the global hypertext is either at a very local level (within individual documents or in closely clustered groups of documents) or artificially superimposed as an organisational convenience (navigational shortcuts from one site to another).
One feature of the Web is the mixture of co-operative and autonomous components of its use. The organisation of each Web site is independent of any of the others, but the details of this organisation, and summaries of the available data are shared co-operatively with other organisations, and frequently published by `key' sites to the benefit of everyone. In contrast, the authoring of the documents at each site is typically performed in isolation, and without reference to the documents available elsewhere on the network. A co-operative effort may be involved within a site to make its collection of documents coherent, but this breaks down at the larger scale and is not exhibited between sites. In other words, according to [Stotts91], at a certain scale the Web ceases to be a hyperdocument (no authored intent and no coherence) and becomes a hyperbase.
The point at which this transformation occurs is the point at which the Web as a hypertext becomes difficult for a reader to use. Accordingly a key feature of a hyperbase is the need to supplement link following with data querying as an information discovery strategy. Up to recently, with the patchy coverage of most search engines, this was not a viable option. However the advent of more sophisticated Web searchers (such as Digital's Alta Vista service) has made this strategy much more useful.
The exact scale at which the transformation between hyperdocument and hyperbase occurs is not fixed: it is certainly possible to store a collection of mainly unrelated articles as a single resource (or even a single document): this is a hyperbase at a very localised scale. Conversely, it should be possible to author a document which draws together information in resources across the global network, from widely diverse sites: this is a hyperdocument (see Figure 2) at the global scale. The facilities of the DLS, which allow an author to create links between arbitrary Web documents and then to publish those links for the benefit of other users, provide a basis for producing such a global hyperdocument.
Experience with extrapolating local information collections onto a distributed platform such as the Web [De Roure95] shows that a clearly structured information design based around the aggregation of resources is highly useful. Collecting individual documents and multimedia information 'assets' into resources allows a degree of abstraction and helps to promote information re-use [Davis93]. A resource may collect together not only documents, but also linkbases suitable to help browsing the documents. The linkbases may be classified as internal to overlay a particular structure onto the resource and define a user's navigation through the assets contained in the resource, or as external to take readers to related materials in other resources, or to bring readers from other resources into this one.
Collections of these resources are considered hypermedia applications [Goose96], the published information environment. (The use of the word application in this context is unfortunate, but historical.) For example, a Biology application may consist of three resources:
In turn, the biology dictionary consists of the text of the dictionary along with links that point to each entry from any occurrence of that entry's headword in any document (an external linkbase pointing into the document). The textbook consists of the documents which contain its various chapters along with an internal linkbase that joins the chapters together and provides a table of contents and index, as well as external linkbases that point to specific online resources (videos, databases, pictures, journals) and point into the book's key passages from any use of appropriate key phrases by any documents. Similarly the lecture notes consist of various documents joined by an internal linkbase, together with an external linkbase to tie the notes to relevant passages in the book, and an external linkbase to provide a specific, highly tailored path through the textbook (i.e. a linkbase in one resource that applies to another resource).
Taken together, these parts form a whole whose coherence is derived from the author's strategy partly in creating documents (lecture notes), partly in defining suitable links and partly in identifying suitable material to join together. The 'whole' is capable of being extended still further by having extra resources added such as student essays or quizes, each of which would be retrospectively linked into the existing corpus by the existing links. [Hall95] discusses an example of this technique in a non-distributed context, but note that in this example, the documents that compose the dictionary and textbook may well not be available at the same site as the lecture notes and links.
This example clearly demonstrates the need to keep track of the components of the various resources being used, especially where resources may be added in a modular, plug-and-play fashion. The Document Management System (whose user interface is shown in figure 3) is a fulfilment of this requirement. (The name is historical and unfortunate because documents are not managed in the sense of being controlled.) The DMS deals with docuverses (a term coined by Ted Nelson [Nelson87] to refer to the universe of documents, but used here specifically to refer to the universe of documents with respect to a particular resource) which hold references to documents, linkbases and other docuverses. Figure 3 shows a docuverse (the 'root') which contains docuverses relating to various aspects of our home city, and a set of documents about the shopping facilities; the user interface is modelled on the familiar File Manager metaphor from Microsoft Windows 3 operating system. When the user starts to browse a docuverse, all the documents referenced by that docuverse are displayed on the right, together with any docuverses referenced on the left. A docuverse is represented by an open folder icon if it is either the 'current docuverse' (i.e. its documents are being shown on the right) or if it is an 'ancestor' of the current docuverse (the equivalent of being 'on the path' in the File Manager metaphor). Clicking on any docuverse makes it the current docuverse, and lists its files on the right of the screen. Clicking on any of the document icons retrieves that document from the Web and displays it in the current Web viewer.
All of these operations are performed with respect to an arbitrary 'root' docuverse, from which all the operations start. The root is probably associated with the user and served from that user's home machine, but any subsequent docuverse may (and probably will) reside on a different machine from its ancestors. Each resource is described by a docuverse, and each application is described by a docuverse that points at many resource docuverses.
In fact a docuverse is conceptually a kind of linkbase which links documents to metadocuments (i.e. document contents to information about that document, such as the author, title, abstract, keywords, version number, date of publication, etc). This is information that would often be directly included in the document contents (such as a TEI header [Burnard94]), but allowing it to be maintained separately yields the same benefits as separate links (see [Hall 94 for an exposition of these). For reasons of efficiency the metadata is included in the link itself, otherwise network latency would render the docuverse unelaboratable. Figure 4 shows an extract of the SGML coding of one of the docuverses seen displayed in figure 3; each entry contains a URL for the document data, together with a set of tags that describe that data.
<document> <URL>http://journals.ecs.soton.ac.uk/ <title>Open Journal Project <keywords>electronic publishing, electronic libraries, hypertext <type>HTML<time-stamp>Tue Aug 29 11:53:40 BST 1995 <document> <URL>http://bedrock.ecs.soton.ac.uk/Microcosm/papers.html <title>Microcosm Literature <keywords>hypertext, open hypermedia <type>HTML<time-stamp>Mon Jan 11 12:26:32 GMT 1994 <document> <title>All About Southampton <URL>http://journals.ecs.soton.ac.uk/dms/soton.docu <keywords>Southampton, leisure, geography <type>docuverse<time-stamp>Mon Jan 29 11:29:13 GMT 1996
A docuverse can be built manually (by constructing the marked-up file from lists of URLs and previously-obtained meta data), or by using a DMS proxy-browser facility that allows the user to navigate the Web under DMS control, with a set of buttons added in a frame at the top of each page to allow information about the currently-viewed page to be added to, deleted from or altered in the currently open docuverse.
The DMS is a mechanism to support post hoc integration of online multimedia information assets via the DLS, and is useful for specifying, visualising and building distributed information applications. It is currently under development and at an early stage of implementation, but initial trials seem to suggest that it will prove increasingly valuable as its user interface improves.
The DMS provides similar services to the bookmark facilities of many Web browsers. The difference is that the lists are distributed across many sites and intended to be shared by many users. Similar (but more developed) work is being done by other sections of the Internet community, for example the use of IAFA templates for resource description, or WHOIS++ for resource discovery (see [Hiom95] for a description of these technologies). Where the DMS approach outlined here differs is that each resource description unit (the docuverse) is intended to be used as a building block in constructing a large, distributed information application.
The combination of the DLS and the DMS creates a very open authoring environment: links can be added to material which the author has no direct control over, and so a wide range of generic information resources can be aggregated together in a reasonably coherent whole. (The issue of creating links to subscription materials has not so far been mentioned, but there is no implication of copyright theft in the use of the DLS or DMS. Just because an author can access a work to create a link to/from it does not mean that any reader can subsequently gain access to the document by following the link. The DLS provides the address of the endpoint of a link, not the resource itself, so that traversing the link requires the end user's identity to be checked as usual by the host of the link endpoint.)
The much larger problem, mentioned previously, is that of resource consistency. If an author is collecting information (meta-data or link information) about documents that they have no control over, then the information may decay over time, breaking the links or rendering the DMS browser useless. Other systems provide environments which make this impossible, and we focus on Hyper-G [Andrews95], a well-known example of such system, in this section.
In contrast to the DMS, one of the main characteristics of Hyper-G [Flohr95] is its guarantee of consistency: its undertaking to keep strict track of all documents and interdocument hypertext links which it handles. Hyper-G has a superficially similar architecture to the World Wide Web: client browsers are provided documents by network servers, but like the DLS and unlike the Web the hypertext links are stored independently. Hyper-G provides support for link maintenance and management, linking between different media types, different sets of links for different users, a docuverse, text retrieval and some visualisation tools for navigating around 'clusters' of related materials.
Each Hyper-G server maintains its own document management system, which keeps the attributes of the documents on the server, a link database which maintains the links, and an information retrieval engine, which can retrieve on both the attributes of the document and also the full text content of the document. The servers themselves may be arranged into hierarchies underneath a world wide 'root' server, but the user connects directly to only one server. Hyper-G can also arrange to collect documents from other servers such as Web and Gopher servers.
The Hyper-G client browsers provide an interface for document and catalogue browsing, authoring and link creation, supporting a variety of standard text, picture, movie and 3D data formats.
Both within documents and between documents hypertext integrity is maintained by the authoring clients. Each document knows the id's of all the links it uses, and even though they are stored externally when a client loads a document it is also able to load all the links it requires. The client is then able to edit the document (or move it or delete it) without causing integrity problems, since at the client end all links are effectively embedded within the document.
The advantage then of the Hyper-G service (or of any hyperbase service), is that it forms an enclosed world of documents: its docuverses contain not just of references to the documents and links, but the documents and links themselves. The disadvantages of this service are that the documents have to be imported into the environment from their development environment, that they have to be translated into a format for which a Hyper-G viewer has been written, and, more generally, that they have to be owned by the database, i.e. brought into the author's domain of authority (in other words you can't link things you don't own).
As we have explained previously, the DMS stores pointers to the documents it is aware (along with the document meta-data). This approach is more fragile than the hyperbase approach, in which the system stores the document itself, as there is always the possibility that the owner of the document might move it or remove it, causing the pointer to dangle. The advantage of this approach is that it enables one to keep references to other peoples' data, without needing to have ownership of that data.
Similarly, DLS links consist of references into objects within documents, rather than marked up links. These links may be fragile if the content of the document changes, leaving the pointer referencing the wrong object within the data. This is known as the editing problem. However, the advantage of this system is that one can make links in other peoples' data. The fragility of this system may seem at first sight to be an intolerable problem. Of course we are all familiar with the problem of dangling links in the Web, but the situation with the DMS is no worse. In fact it can be considerably better.
The advantage of the DMS is that it creates a closure of all the documents we are interested in, and provides an explicit list of those resources. This makes it very easy to produce tools which can iterate over the documents in the list: such tools can then regularly check that each server and document known to the DMS is still accessible, and can check, using the HEAD command of the HTTP protocol, that the document has not modified. The DMS may be used to indicate if there has been a problem accessing a document, for example by colouring the document icon. If the document continues to be inaccessible, its record in the DMS could simply be garbage-collected.
A further possibility, using the DMS, is to store a local copy of any document that is referenced. If the Web document then became inaccessible we could offer the user the opportunity to view the locally stored copy. This version may not be as up-to-date as the remote copy, but might still give the user the information that was required. Alternatively, if the referenced document is updated, then the DMS is capable of displaying a summary of changes, if the user is interested.
The situation with links in the DLS is also better than it might appear at first. The DLS does not mark-up link anchors within the text. The start anchor of DLS links is defined by the text string from which the link can be followed, and optionally by the name of the document. Thus a link might be available from any occurrence of a particular text string ( a generic link) or from any occurrence of that string within a particular document (a local link). Such anchors are not affected by changing the position of the string(s) within the document. The end anchor of the DLS link is a point within a document, and the intended position of this point might be affected if the document is altered.
There are a number of ways of dealing with this. We might make links only to the top of documents or we might express links in terms of the first occurrence of some string, rather than in terms of a position within a file. In any case, the DLS is always able to warn the user if the document has changes (and thus that the end of the link may be wrong) since it keeps the date that the document was last modified at the time that the link was made, as information inside the link. If at some subsequent traversal of the link the DLS observes that the date is no longer correct, it can warn the user. Again, because the links are stored explicitly, we can build tools to iterate over the links and discover which links dangle, or point to documents that have changed, so that dangling links may be garbage collected, and users may be warned of links with suspect pointers.
By taking advantage of the docuverses we can compensate for some of the pitfalls of the open model. These strategies (currently under development) provide a way of adding varying degrees of 'closedness' for applications that require some form of guarantee of the availability of data. This guarantee is in the form of post-hoc resource checking, but could be extended to provide agent-based pre-emptive polling strategies.
One of the major projects which is using DLS and DMS is the Open Journals Framework funded in the UK by the FIGIT initiative of the Electronic Libraries programme (see <URL: http://journals.ecs.soton.ac.uk> for more information on this project). The aim of the project is to address some issues of the digital information environment which is currently being formed in academic libraries as an increasing proportion of their information assets are becoming electronic, available on CD-ROM or via the Internet.
A problem for users of library information services in Higher Education is the isolated and diverse nature of the electronic information resources. Although a user can (in theory) from the same terminal access many dozens of journals, databases and articles on subjects of interest, it is necessary to navigate a complicated path through many providers information gateways in order to locate any particular piece of information of (as yet) undetermined relevance.
The goal of the project is to develop a framework of information retrieval technologies and electronic publishing practises to be used by information providers (especially journal publishers) which will allow them to make their publications available not as isolated, one-off resources, but as co-operating assets within an information delivery environment such as a library at an institution of Higher Education. This goal is to be achieved by using the DLS and DMS to seamlessly integrate journals that are available electronically over the network with other journals and information resources that are also available on the network.
The result of the first part of the project is to produce a demonstrator of the capabilities of the DLS and DMS: it is a more elaborate version of the biology application explained above, and consists of articles from a number of biology journals, online glossaries, protein databases, bibliographic databases and lecture notes. All these resources, some of which are owned by the project, and some of which are generic resources available on the Web, are linked together by DLS linkbases and presented using the DMS interface.
In summary we have seen that the 'closed' approach can provide hypermedia services with guarantees of data consistency inside a sealed information universe for a limited set of information types, whereas the 'open' approach can provide extensible hypermedia services for documents of any type, maintained within any environment but without any hard and fast guarantees of data consistency. It may be tempting to draw the conclusion that the former approach is suitable for real-world, mission-critical applications, whereas the latter is only acceptable for unimportant information.
However, it is worth considering real world examples before jumping to this conclusion. Although a library is often considered to be a pre-eminent example of a controlled, closed information environment, [Levy95] challenges this description, since real world collections are subject to 'crumble', i.e. decay over time. Hence library catalogues (as well as the documents they describe) require constant and active maintenance, without which their consistency cannot be guaranteed. It is also the case that as information consumers in the real world we deal with uncertainty of access all the time: if the host site of some data is down, or there are network problems that stop us from reaching the site, (or indeed if the library is closed or a book has been borrowed) then we have heuristics that help us to take appropriate action and to continue our tasks. If we take this point of view then we may decide to work with any potential uncertainty, especially in order to gain the benefits of an open environment, and to develop futher safeguards for an open environment by implementing constant docuverse monitoring, or designing a way of declaring exception handling facilities within a docuverse.
Finally, let us mention two user groups on the Web. Commercial users who maintain a WWW site to distribute their own company and product information are likely to benefit from a closed, hyperbase-style system that is especially suitable for single-site hypertexts. In contrast, the academic community has been impacted by the online publishing phenomenon in a different way ([Harnad90]). Consequently online scholarly communities are forming that require shared, distributed publishing facilities where an open model would excel.
This work was partially supported by JISC grant ELP2/35 and a UK ROPA award.
is a Senior Research Fellow in the University of
Southamptons Multimedia Research Group. His research interests
include the application of HyTime, SGML and DSSSL to hypermedia
applications, in particular WWW, and he menages the Open Journal
Framework project and the ongoing development of the Distributed
No biographical information available.
David De Roure
is a lecturer in Computer Science at the
University of Southampton, UK, researching into distributed
systems and in particular addressing the scalability and
complexity of networked information systems. David leads the
Distributed Systems and Multimedia Networking activities in the
Multimedia Research Group.
is a Professor of Computer Science at the University
of Southampton. She is variously a Director of the Multimedia
Research Group, the Universitys Interactive Learning Centre and
the Digital Library Centre, researching into multimedia
information systems and their application to industry, commerce
is a lecturer in Computer Science at the University
of Southampton, UK, and was a founder member of the multimedia
research group. He was one of the inventors of the Microcosm open
hypermedia system, and is manager of the Microcosm research
laboratory. His research interests include data integrity in open
hypermedia systems and the application of multimedia information
retrieval techniques to corporate information systems and to