Assoc. Prof. Liddy Nevile, Principal Researcher, Melbourne IT Ltd, Level 2 120 King Street, Melbourne 3000, Victoria, Australia. liddy@melbourneit.com.au
Dublin Core; RDF; Education; Discovery
This paper describes the design of the terms and syntax developed for the Victorian Department of Education, Employment and Training's 'Education Channel' catalog. In particular, the rationale for breaking from what was at the time common extended Dublin Core schemata and the early adoption of a qualified Dublin Core approach is considered. The tools to support the new approach are also described briefly. The Victorian Education Channel is due for release for public access in 2000.
In the late nineteen nineties, the Victorian Government commenced a rigourous programme of improving public access to Government information and services via the Internet. The programme involves access to government available through a series of websites referred to as information 'channels', as well as the provision of a central governemnt website providing links to separate channels and to be searchable in such a way that resources from any departmental sites, relevant to visitors to a particular government site, could find what they needed. These 'channels' have, in their own way, represented the state-of-the-art technologies as they have emerged. The Victorian Education Channel is no exception.
This paper describes the design of the terms and syntax developed for the Victorian Department of Employment, Education and Training's 'Education Channel' catalogue. In particular, the rationale for breaking from what was at the time common extended Dublin Core schemata and the early adoption of a qualified Dublin Core approach is considered. The tools to support the new approach are also described briefly. The Victorian Education Channel is due for release for public access in mid 2000.
The architecture, metadata standard and applications are standards-based and suitable for application in educational settings other than the Victorian DEET system. National adoption (generalisation) of the mapping techniques and processes, while maintaining local control of the resources, provides a potential fast growth path for discovery and retrieval of Australian educational resources and resources for Australian education.
In 1999, the State of Victoria continued its practice of developing 'channels' as atlases of government related materials and services. Already there was a health channel, a land channel, a business channel and an education channel was long awaited. DEET (1) had developed a significant web presence over the previous four years, including a schools website, which is one of the most heavily used education sites in the Australia; a extensive post compulsory education and training website; an on-line delivery system for TAFE courses, and an intranet designed to significantly improve business processes. The provision of a single channel uniting these content sources was an obvious next step in optimising access to key audiences (eg. students, parents, teachers, community).
Two hundred business units within the Department had placed materials or access to services on the web. Hundreds of groups within departmental responsibility, such as schools, published websites with some resources of significant interest to practising teachers and students. The range of media types, audiences, levels of web expertise and interests of the audiences for all these resources and services made it very difficult for both departmental publishers and users to ensure discovery and retrieval of the relevant resources without the noise of many irrelevant resources.
At Victorian Government level, there was pressure to develop a cross-departmental mapping of individual departmental resources in such a way as to make it possible for web users to approach what would appear as a seamless government, to provide a 'whole of government' approach. Users were expected to be able to think in their personal context and discourse, for instance that they were about to move house, and to find that they could access all the services and materials of relevance to them on the web without having to know which government agency, or which level of government, provided the resource they needed. MultiMedia Victoria promoted the idea that users should be supported to look for what were called 'life events', for example, even if government agencies worked to a different taxonomy.
In the real world, the many departments of a government struggle with problems related to which department handles what aspect of any given situation. A single event can trigger the use of many documents, sometimes copies of the same document, within a number of different departments. Julia Schofield (2) introduced the Victorian Government to the concept of a website user's 'life events' as a unifying taxonomy for all government activity. In 1999, there was a strong expectation that such a taxonomy could, in some way to be determined, be built into web access points, such as the channels (see eg 'maxi' [HREF1]), to hide the real 'silo' nature of government activity. Users from within government, as well as the general public, were expected to be able to access resources more efficiently if they were classified according to this taxonomy. In developing the taxonomy that would underlie browsing by the user of the Victorian Health Channel [HREF2], the author tried to find ways of representing health topics so they would be discoverable from a taxonomy of life events. The concept was not well-defined and the author and local originator, Julia Schofield, decided that there was significant research to be done before it would be easy to apply this concept although some use of it had been made in the 'maxi' [HREF3] implementation. The author developed a simple theoretical model that defined a life event as the transition event that triggered the change from one 'life stage' to another and then used life stages in a rough taxonomy. This unsatisfactory experience, and the fact that other channels had not managed to implement the 'life stages' taxonomy, led to its abandonment for the education channel.
The National Australia Archives (NAA [HREF4]), not a Victorian organisation, had taken a different approach and a comprehensive survey of government functions, as related to resources, had been undertaken. This led to the development of a inclusive Australian government functional taxonomy for resources, broad enough to be responsive to natural language. In December 1999, the NAA launched the Australian Government Information and Functions Thesaurus (AGIFT HREF5) consisting of more than 330,000 terms.
The eclectic nature of educational resources, and the multiplicity of roles they might play in people's lives, makes them particularly hard to categorise. A single resource might be wanted by a person who, on different occasions, is acting as a teacher, a graduate student, a parent or a mature age student. Fortunately, classification of educational resources is not a new activity and there are a number of precedents for classification of both physical resources and digital web-based ones. It is also usually the case that the different communities or representatives of people in different roles, classify the same resource differently. This suggested to the Victoria Education channel team that multiple records should be allowed for a single resource, and they should be able to reference different taxonomies.
In some cases, it has been possible for the Victorian Education Channel to reference the work of others, for example to use the thesaurus of the NAA as a controlled vocabulary, and to draw upon the work already done by such organisations as Education.au [HREF6]), the Australian federal association of state educational systems concerned with educational resource discovery; the University of Syracuse (USA)'s Gateway to Educational Materials (GEM [HREF7]); the work associated with the Institute of Electrical and Electronic Engineers (IEEE [HREF8]), the emerging Instructional Management Standard and Learning Objects Metadata (LOM [HREF9]) standards, Ariadne's standard [HREF10] and those of the many other educational communities that had already tried to classify their web resources.
While the problem of classifying educational digital resources for the Victorian Government benefited from the semantic agility of many of these different communities, they did not provide a universal, or global, normalised, taxonomy that could be used to relate all resources to all others. Following the report of the Desire project [HREF11] among others, the author was searching for such a taxonomy and fell back to investigating the more classical library taxonomies such as Library of Congress Subject Headings (LCSH [HREF12]) and the Dewey Decimal Classification (DDC [HREF13]) system.
In the end, DDC was chosen over local taxonomies and LCSH because it appeared that the best putative (3) classification tools were being developed for DDC and that these would be accessible to the Victorian Education Channel. The Dewey taxonomy was chosen also because even if it was normally used to determine where a book should be placed on a shelf in a library, it was proven as a system that managed to cover everything. It was also relevant that educational libraries in Victoria tended to have been organised by reference to the DDC and so the taxonomy would be familiar to university trained practising teachers and student teachers.
Locally valued taxonomies, such as the Curriculum Corporation's Schools Catalogue Information Service (SCIS [HREF14]), could be accommodated as fixed vocabularies from which values of classification elements could be chosen, and to which reference could be available for discovery purposes. In effect, the semantics and the structure of the classification could be specified so that any community within the broad educational category could work with locally comprehensible vocabularies and taxonomies, while supporting a more global approach to discovery of their resources.
The global taxonomy, DDC, was used to provide a unifying base set of values for the classification criterion 'subject'. Expressed as values of dc.subject and defined by the DCMI (4), values selected from the Dewey system would be used to provide a context for the metadata record that contained it. The advantage of this approach is that even if discovery is not undertaken by reference to a Dewey term, the metadata record is related by the Dewey value to all other, similarly classified metadata records. When a user is searching and is getting closer to her target, the option to browse rather than work through search terms can be organised by the Dewey subject value and associated subject path. The subject values are part of an object oriented system, and if a different global taxonomy were to be desired some time in the future, the file could be replaced without breaking the system. The following sections aim to explain the approach that was adopted with respect to classification criteria or elements.
The Dublin Core (DC [HREF15]) provides 15 'core' elements that, if used to describe a web resource, have been considered appropriate for basic discovery purposes. The choice resulted from extensive consultation and harmonisation activities undertaken by the DCMI and constantly in review. In some particular circumstances, values for these elements are considered too gross to be useful. In such cases, two ways of qualifying the values have been adopted (beyond the third approach (5) which is to just add more, non-DC elements to the list). First, sub-elements can be used to qualify the meaning of an element so long as the qualification of the element does not take the relationship of the value outside the meaning of the element but merely refines it (6). Secondly, understanding of the value itself might need to be clarified and provision of controlled vocabularies, definitions or formats can assist in this process.
For example, if the date associated with a web resource is 12-1-99, it may not be clear whether this is a date in American or Australian notation, or whether this is the date of creation or modification of the web resource. The DC element date makes some sense with the value 12-1-99 but more sense if it is possible to determine that this is to be read as 12th January and not 1st December, 1999. Qualification of the value of the element, the 12-1-99, is easily achieved by allowing a reference to the value format being used, such as DC.date.dd-mm-yy (7). Dublin Core rules require users of such formats to identify and publish the formats applied for the benefit of users. DC.date.created.dd-mm-yy allows the user to know that the date given is the date of creation of the resource. This qualification of the element does not break the meaning of the element but, instead, enhances it. A sub-element was only to be a sub-element (element qualifier) if it satisfied the 'dumb-down' rule (8), that is, its value could be read meaningfully as a value of the element.
In the first few years of experimental use of DC, most agencies used the 15 DC elements with no qualification (9) and added extra elements where they wanted precision. In particular, this approach fitted with the use of HTML, the syntax used by most to express DC information. HTML is not structured in a way that allows for complex statements to be constructed and read appropriately, particularly not complex statements such as the DC elements with embedded element and value qualifiers.
The World Wide Web Consortium (W3C [HREF16]), the body responsible for the HTML standard, already had another standard that provided for transmission of information to be associated with a web resource even if it did not emanate from the same publisher or source as the web resource it described. The Platform for Internet Content Selection (PICS [HREF17]) offered a model for resources and metadata to be separated. Not until this standard was upgraded to include structure, repetition and strings could it be used for complex DC statements (10).
The new Resource Description Framework (RDF) needs to be stable and published if DC metadata expressed in it is to be interoperable and thus generally useful for resource discovery. Currently, there are two recommended RDF syntaxes, a recommended data model, and a lot of controversy about the role of schema and the relationship between XML and RDF schema. Work is ongoing for the W3C RDF Working Group that is wrestling with these problems and the DCMI working groups that deal with the syntactical problems of the DC standard. The hope of many is that these difficulties will be resolved soon and the outcome will be a single standard to support the 'semantic web', now sought by many (11).
The Victorian Education Channel project decided to preempt such developments.
Qualified Dublin Core was now being developed to replace the earlier standard. It was considered to be a significant improvement on the earlier form, particularly as it would increase yet again the levels of compexity that could be contained in the metadata records. RDF would reduce the payload attached to this increase in expression complexity by providing for the relationship of one record or record's entry with another. For example, a record might refer to another record to provide information about an author, rather than reproduce the information contained in that other record.
The richness of qualified DC would be expressed and used by the applications being designed for the Channel and it was to be assumed that others would develop such facilities, based on the same, or a slightly modified version of DC. The Channel would be classifying resources and maintaining that information separately from the resources which would normally reside on the developers' web server. Whatever agency supplied the metadata, it would be rendered in RDF and used in that form by the Channel applications. It may also be made available for export from the Channel in RDF. All metadata in the Channel repository would probably be governed by a single, publicly-visible standard such as the Victorian Education Metadata Standard (VEMS [HREF18]).
VEMS included qualified Dublin Core at a time when others were still working with standard Dublin Core and achieving their goals by adding extensions. The most organised and comprehensive of these groups was the group concerned with the ability to link and thus combine educational resources, particularly those that provided instruction with those that reported assessment. This goes beyond the discovery and retrieval concerns of the DC. The IMS Global Learning Consortium's standard (IMS [HREF19]), at the time included as many as 42 elements for describing a single resource compared to 15 for Dublin Core. In fact, other educational standards based on Dublin Core and confined to discovery and retrieval also had more than 15 elements, but usually not many more.
The problem of getting the standard implemented was considered to be as challenging as the determination of a standard. Being able to ask for values for 15 elements, rather than 20 or 30, was attractive from a marketing perspective and this was important. Completing forms is not everyone's favourite activity, sufficient numbers of cataloguers could not be paid and thus compelled to undertake this activity, and so the whole system would depend upon the good graces of those who were willing and able to implement the standard.
Two priorities were recognised: gaining wide acceptance of the standard, which would only come about if there was wide involvement in its development, and simplicity, making life as easy as possible for implementers. Tools would be built to automate as much of the cataloguing process as possible but these could not ever be expected to replace all the human input required and could not be expected to be as accurate in some respects. Computers might copy titles accurately but they would not be as good as humans as assessing more subjective classification aspects.
Departmental publishing systems would be designed to generate metadata as part of the publishing process. This would be then be ported to the channel. Updating of the metadata would become a part of content maintenance by DEET staff and partners by being built into web publishing and quality assurance processes.
'Fifteen elements', even if they were qualified and that meant more, sounded good. Fifteen, if dumbed-down, would be standard and make the system more interoperable with other systems; fifteen would test the syntax of RDF and provide an interesting challenge for the developers. Fifteen it would be, if they made sense and could be justified. (In fact, the records containing the fifteen metadata elements were themselves to be described by metadata, and for this a meta-metadata standard was adopted.) Given its relationship to the Victorian Government and the international educational world, it was essential for implementation purposes that the VEMS would include all that was required to assure EdNA and AGLS compliance, and be appropriate for interchange with metadata that was compliant with other important international standards, in particular IMS. In fact, the channel would provide a tool through which collaboration with other knowledge discovery systems (e.g. EdNA Online) could be assured (by importing and and exporting RDF).
The VEMS that defined the implementation of DC for the Channel was important, but organisation of the resources for discovery purposes requires more than classification. Searchers need to have pathways into the possibilities that make sense to them, whatever was good for the cataloguers. A guiding principle, of great support, was the assertion by one experienced expert in the field that it may be useful to think of the cataloguer as having a conversation with the discoverer, just that the conversation is not taking place in real time. "What would the cataloguer want to tell the discoverer?" became the driving question for the definition of the classification system.
DC does not help with the vocabulary of resource description beyond providing a structure for its inclusion. Users trying to discover a resource can be expected to have compound interests, such as the teacher who wants resources to help with teaching small children about ecology when she finds a frog in her garden. The teacher may be familiar with the curriculum structure for her school, she may know that frogs need careful treatment but still not be sure what that means. She may not be sure what young children can understand in terms of ecological principles or which of the frog-based activities will help them engage with the principles. The teacher may not know much about frogs. How will a user looking for frogs, and curriculum guidance, and ecological principles, and help with young children's intuitive understanding, use the Victorian Education Channel?
The 15 DC elements allow for information about the resource with the information for finding resources with the relevant subject content contained within a few only of the 15: DC.subject, DC.description, and possibly DC.coverage. DC.description is used for a text description of the resource and, in the case of the Victorian Channel, constrained to 50 words. DC.subject has a qualified version, DC.subject.keywords that are freely chosen terms suitable for inclusion in the head of the resource for collection by public search engines (as tags). DC.subject.csf are terms chosen from a selective vocabulary developed to describe the newly revised Victorian Curriculum Standards Framework (CSF). But frogs are not necessarily identified in the curriculum statements; they are not the objects of study so much as the vehicle for it.
The overall classification of the resources for the Channel is managed, in fact, by a series of overlapping taxonomies, each serving different purposes. Everything is ordered by Dewey numbers (see above). Although this at first seems odd (given that web resources are not like books and do not need to belong to a linear classification taxonomy because their location is not linear) there is still the problem of how discovery can work from the general to the particular if there is not some hierarchical structure to steer users towards what they want.
The Channel's search applications provide for users to either browse or generate searches from where ever they are in the discovery process. A unifying hierarchy is essential for such an activity. A global taxonomy provides this.
With Dewey for the global taxonomy, the need for a local taxonomy still remained. Different education sectors within the State think of the same object differently: a course or unit in physics at a university might be described as a science course whereas the same course in a vocational sector might be considered as an engineering skill-development course. Local taxonomies can account for differences in discourse, perspective, language, and more.
The problem of having several local taxonomies that overlap, possibly each being used to describe a sub-set of the total resources, is that users of one local taxonomy might miss resources not described using that taxonomy even though the resources are relevant. For this problem, there are two possible solutions: mappings from one taxonomy to another or multiple catalogue records, each using a selective taxonomy.
Generally, there are situations in which particular resources are identified as being of interest to a community and there is a requirement to make them available without the clutter of other similar but irrelevant resources. The specially selected resources should be shown to the user during the discovery process. Such users of pages might need pre-structured searches to take place when they make top-level selections of search terms, sparing them the difficulty of specifying complex searches. These users might need to be offered mappings of taxonomies rather than the original terms of the taxonomies, as would be the case for young children searching through terms that are too sophisticated for their use.
For all these problems that were approached by the provision of levels of taxonomies, there was still a need for adjustment of terminology. In some states of Australia a child of ten years will attend a school and be in Year 5 while in another state the same child would be in Grade 5. Educational thesauri have been developed in the past for indexing of curriculum and educational research materials and these, in their recently up-dated form, provide excellent thesauri for the Channel. They are differentiated from taxonomies as they do not, in the Channel, structure the metadata but rather interpret the terminology.
The Victorian Education Channel is designed to use tiers of taxonomies and thesauri, as shown below.
In this diagram, the surrounding cloud represents the presence of thesauri such as AGIFT and the thesaurus for the Australian Education Index developed by the Australian Council for Educational Research (ACER [HREF20]) providing coverage of all terms for the discoverer. All resources are contained in the full-text index, most have DDC numbers, some have coverage from a specified classification scheme and others, including some with CSF classification (and some are identified by specially-constructed, manually linked mappings). The user, while discovering a resource, selects 'filters' through which to search and view the resource catalogue records. Global and local thesauri combine behind the scenes to form comprehensive but not redundant layers of inclusion of resources.
The use of the term thesaurus has been limited to represent sets of terms that are used by the system to give broader or more specific or alternative meaning, according to a formal mapping between terms, and by the user in as much as they tolerate informal terminology. The dumb-down rule of Dublin Core has been combined with the structured layering of terminology to ensure that the user gets a fully-optioned opportunity for discovery.
The various thesauri and taxonomies are associated with DC element values according to the DC rules that prescribe identification of controlled vocabularies or other value qualifiers. As the relevant controlled vocabularies are generally public and likely to be shared with other agencies for classification purposes, identification is achieved by the addition of an identifying link to the element or sub-element for which the value is being provided. An example might be: DC.subject="TheArts.Dance,Ballet.Level2.Exercises" where scheme="http://csf.somewhere.edu.au" and CSF is the classification taxonomy for Victorian school curriculum materials, available on the web at http://csf.somewhere.edu.au.
The VEMS requirements, as adopted for the first version of the Victorian Education Channel, were used as requirements for the Cataloguers' Workbench, a set of applications that would help those developing metadata for educational resources with the aim of having this metadata incorporated into the Victorian Education Channel [HREF21]. The architecture of the Channel separates the creation and maintenance of metadata from the searching and discovery services provided by the Channel. DEET business units and affiliated organisations contribute metadata to the Channel and then, according to their needs, draw upon the metadata or discovery tools within their own websites or point users to the Channel.
The Cataloguers' Workbench is a suite of applications that produces RDF metadata, fully integrated for presentation to the user via a web interface. The metadata is intended to be stored and searched by the Channel's own applications but can be exported for use by other agencies, including EdNA. The RDF is standard and the tools are modular. In addition to generation of the fifteen DC elements and sub-elements, they generate a full-text index of any resource.
Not all metadata is created by the user: some of the applications generate what has been called 'putative' metadata. Determining which Dewey Decimal Classification (DDC) number applies to a particular resource is a skilled activity, traditionally undertaken by trained cataloguers. The Cataloguers' Workbench has a DDC generator that can 'read' the resource content and propose some DDC numbers. This metadata is designed to be offered to the cataloguer for verification. Recognition of a misfit is assumed to be a lot quicker and easier than selection of a number from the several hundred thousand DDC numbers available. The human metadata creator is to be assisted by a putative metadata generator that in some cases might be able to determine better than the human what a value should be: e.g. the size of a file, but in other cases, will need to be checked by a human, eg the DDC number.
The metadata that is not available from document management services will be created by cataloguers (so-named in recognition of their role if not their expertise). In many cases, the users are expected to be amateur cataloguers who have expertise in their educational work and are likely to know what terms those who seek the resources can be expected to use when trying to find them. Again, experienced, expert advice was that well-intentioned cataloguers do a very good job of anticipating the discoverers' requirements and there are many more potentially well-intentioned cataloguers than professional cataloguers.
What has resulted from the work for the Victorian Education Channel is a Victorian Education Metadata Standard (VEMS) that has, in its own way, broken new ground by adopting a new approach to the problem. At the time of writing, the VEMS has catalysed activity (12) at both the federal and international level. The Dublin Core Metadata Initiative (DCMI) has formed an international DC-education Working Group charged with the task of determining a recommendation for the DCMI for sub-elements and element qualifiers for educational use of DC. It is not clear at the time of writing whether this process will result in 15, straight but qualified, DC elements or if there will be additional elements recognised as necessary for even minimal, interoperable metadata about educational resources.
The Victorian education department, DEET, has undertaken a major re-development of its handling of resources destined for publication via the web, particularly providing for cataloguing and searching based on the VEMS. The Department and its associated agencies have a major on-line publication programme to support the new Curriculum and Standards Framework. This includes publishing at least 30,000 pages of teacher support materials for teachers in the year 2000 and will provide full support for the VEMS.
The Victorian Education Web is large and distributed. Tens of thousands of carefully constructed catalogue records, developed during the last few years, will be converted and added to the Channel repository for indexing. A large number of web resources will be classified and discoverable from quality catalogues within a short time of the launching of the Victorian Education Channel. Resources from the State's Museum, Gallery and Library will also become available as a result of the classification of their collections by teachers. A new technology will be on trial and, if nothing else, an opportunity to work with a substantial, highly ordered set of web resources will be available to inform developers for the future. As important as this will be the way in which the various organisations will be able to work together more effectively to support shared objectives such as public education.
[ Full Paper ] [ Presentation ] [ Proceedings ] [ AusWeb2K Home Page ]