Michael Gardner, The University of Essex, mgardner@essex.ac.uk
This position paper claims that there are significant benefits from incorporating semantic web techniques within e-learning systems. A simple description of the use of ontologies is given, followed by three examples of how systems which use this approach can enhance the overall user experience. Finally we summarise the clear benefits and opportunities for the e-learning community.
Contents |
In the field of computer and information science, an ontology can be defined as a data model that represents a domain and is used to reason about the objects in that domain and the relations between them. An ontology will represent the particular meanings of terms as they apply to a domain. Ontologies are a building block for the future semantic web, which is a vision of web resources that are understandable by computers, so that they can search and perform actions in a standardized way.
DELTA is a system which allows distributed resources to be submitted, searched and retrieved, based on standardized meta-data. Extensibility is a key issue, in that it should be possible for new resources to be added and shared by the learning community. This is achieved by providing a managed meta-data server, which contains the schemas for the sharable resources. DELTA also contains a number of subject ontologies which support the user in categorizing and finding new resources. At the heart of the system, all resources are categorized by a pedagogic ontology. This ontology supports practitioners in contextualizing their needs within a set of core pedagogic principles, which will affect their choice of learning resources for any given learning activity. A Delta resource wizard guides the practitioner through the process of classifying a resource against its pedagogical context.
Semantic web technology can significantly improve the effectiveness of digital resource sharing. By using an ontology inference service, searching no longer need be constrained to matching the content only, but also by inferring the true meaning of the concept it is possible to retrieve all knowledge equivalent resources. The current DELTA system incorporates a number of additional subject-domain ontologies which can be used to extend and enhance the search capability within DELTA. The main challenge in constructing an educational subject ontology is in adequately defining its scope and granularity, and how it should be represented to the user.
Subject ontologies can also act as a recall device, where instead of explicitly defining the subjects to be searched, implicit subject resources derived from ontology inferences will also be retrieved. They can improve the relevance of retrieved resources. In this way, the ontology acts as a precision device, improving the relevance of resources retrieved.
There are also clear research issues which arise from this approach particularly in the area of community defined ontologies and the "folksonomy" approach used by such tools as Flickr and Del.icio.us. In these systems a folksonomy of terms or labels that describe online content is generated autonomously by the user population. For example, if I upload my holiday photos to Flickr then I can describe the photos with textual labels such as ‘beach’, or ‘castle’, etc. This then enables other users to find my pictures if they are looking for a particular subject area (eg. castles of Britain). This type of social classification can work well for certain kinds of information, yet raises several issues particularly where a knowledge domain needs to be validated by some kind of information architect.
We are also investigating ways of making e-learning resources available to online communities through the use of social-networking tools. ResourceBrowser addresses this problem by integrating both the DELTA and eProfile (social networking tool also produced by our team) toolkits into a single user-interface. This demonstrator allows users to view and search their social-networks (based on the eProfile functionality), plus they can search resources in Delta. There are clear benefits from linking the two toolkits together. For example, if the user finds a person (in eProfile) who has Delta resources, they can also explore these resources within the same interface. Also, if a user retrieves resources from Delta then they can also explore the resource owners social network. We have also enhanced the Delta search interface by using a graphical user-interface (similar to the eProfile Touchgraph interface) to allow users to select different subject ontologies (from Delta) and browse these in a visual way.
This work makes uses of a standardised meta-data framework called ‘Friend of a Friend’ (FOAF) which provides a standard for defining user profiles and exploring the relationships between users (which is ultimately a type of ontology).
Both manual and automatic construction of meta-data come with their very own problems. The quality of automatically constructed knowledge relies very much on the data it is derived from, whereas manual construction of meta- data is often prohibitively expensive and once constructed such sources can be difficult to maintain.
AUTODISCOVER aims to address this problem by building a toolkit (using existing technologies) that can trawl a user’s PC and automatically construct meta-data for the documents on that desktop. The user will then be able to manually review and modify the concept-map and meta-data descriptions produced by the system, and select which documents they wish to share. The user will then be able to upload (using AUTODISCOVER) some or all of these meta-data descriptions to Delta (using the Delta OAI-PMH interface), which will allow other users to access these documents (the documents themselves will still remain on the user’s desktop PC). In addition the tool will create a domain model for the documents that are contained on the user’s PC, based on the terms from the document collection (nouns and noun phrases). These (concepts) are then organised as a set of simple hierarchies that form the domain model. This domain model will then be uploaded to the user’s profile on eProfile. In this way it will be possible for users on eProfile to explore each other’s domain models, and also explore and retrieve desktop documents found by AUTODISCOVER.
We are still far away from Berners-Lee’s vision of a Semantic Web. Despite the fact that there is a constantly growing quantity of data sources containing explicitly encoded semantics, most data still comes with much less structure, usually encoding presentation details rather than content. The increasing use of meta-data to mark up document collections suffers from a number of problems, e.g. consistency issues, and meta-data developed or tuned for a particular task may be inappropriate for another task.
This paper has given three examples of systems that use a semantic web approach to enhance the user experience. Delta provides a mechanism for getting practitioners to consider the pedagogical context when choosing their resources, and provides an example of how subject ontologies can enhance the search and retrieval processes. The eProfile/ResourceBrowser system extends this concept by situating resources within the context of the social networks of the online communities, and allowing users to explore both the underlying networks of resources and people.
Finally we have identified the problem of automatically indexing information with accurate meta-data. Folksonomies provide one way of doing this by allowing the user community to create their own descriptions and tags. With AUTODISCOVER we described how it is possible to derive meta-data semi- automatically using the documents’ content. With this system the user would be free to accept or reject anything that results from this meta-data construction process. Such a user-controlled knowledge acquisition step combines the advantage of automatic approaches (rapid knowledge extraction) with that of a manual intervention step (to ensure quality of the extracted knowledge). This approach is primarily focused on exploiting the documents’ internal mark-up structure in order to construct domain-specific meta-data automatically. All documents have some internal structure ranging from explicit markup languages to the structure of a document consisting of a title, headings, paragraphs, captions etc. Our actual knowledge extraction approach can be performed automatically with no expert knowledge. It is largely independent of the actual language used in the documents and more importantly it uncovers the structure of a domain of documents while itself being domain-independent. The resulting “conceptual maps” also allow us to explore and share the document collections more easily (this links back directly into the results of our previous work in the eProfile project).
