The Conceptual Open Hypermedia Service (COHSE) system [COHSEa] supports dynamic linking of documents through the use of an ontology/terminology. COHSE's motivation, design, architecture and use have been described elsewhere [COHSEb, COHSEc], but we provide here a brief overview.
COHSE provides an agent that operates on behalf of the users and adds links to documents. The agent can be provided either as a client side implementation, interacting with the browser, or as a proxy, through which all browsing activity takes place. The agent adds links to existing documents on the web. Potential link anchors are identified through one of two mechanisms:
Once a link source has been identified, the agent retrieves link targets and displays these to the user. A variety of alternative mechanisms are available by which a link target can be presented to the user — the current implementation employs linkboxes, which pop up when the user clicks on the link source anchor, and show a list of available targets. Linkboxes may also show definitional or scope notes information relating to the term in its terminology.
A simplified overview of COHSE's architecture is shown below. The agent makes use of two services: An Ontology (or Terminology) Service; and a Resource Manager. The Ontology Service is responsible for providing information relation the conceptual models that are being used as the context for linking. Services include providing lists of terms in an ontology and answering queries regarding relationships between terms (e.g. sub/super class relationships). The Resource Manager is responsible for providing mappings from concepts to resources (used in determining link targets) and resources to concepts (used in determining link sources when employing annotation linking). COHSE's current implementation has concentrated primarily on generic linking.
A key aspect of COHSE is the open-ness of the system and the ability to augment third-party documents with extra links. The use of standardised representations for vocabularies is also crucial, enabling the integration of (third-party) information sources into the system without additional tailoring.
The figures below illustrate COHSE in action, using the Gene Ontology [GO]. The first shows the addition of links to a page (terms are highlighted), with a linkbox relating to the term "cytosol".
In the second, the term "membrane" has been highlighted and selected. In this case, as well as showing relevant targets for the term itself, the structure within the ontology has been used to extend the retrieval, providing targets relating to more specific resources. In this way, the structure in the ontology is used to help control the user's navigation around the linked resources.
To date, the conceptual models used in COHSE have been based on OWL ontologies. The super/subclass hierarchies are then used to drive the navigation process. Although this provides some utility, we do not consider this to be ideal — it is not always the case that super/sub class relationships provide the navigational structures that users require. Existing vocabularies have to be "massaged" into an appropriate format that the agent can then use, but which do not faithfully reflect the intention of the model. Basing the system on vocabularies represented in SKOS would a) make a number of existing vocabularies available for use within the system; and b) provide a better "fit" between the navigational models employed by COHSE and the structure represented in the knowledge models. In many cases, the looser notions of "broader" and "narrower" are more appropriate relationships to base our navigation on rather than the strict sub/superclass provided in OWL.
COHSE accesses terminologies and resources via services. The operations required on these services include:
Two specific use cases under investigation are the linking of documentation within Sun Microsystems, and the linking of resources for bioinformaticians in the SeaLife project.
Sun Microsystems provide a number of different web sites relating to Sun products. These include:
Each of these sites provides information relating to Sun products, but there is no explicit cross-site linking. Sun also provide a Unified Product Taxonomy (UPT). The UPT describes the components, systems and software provided by Sun Microsystems, along with relationships between those components. The UPT also provides lexical information relating to the names (and synonyms) of the products, and can thus be viewed as a thesaurus.
Through the use of COHSE with the UPT, cross-site linking can be achieved, providing links from pages describing products on the SunStore site to FAQs, support information in the SunSolve site or additional internal resources.
The SeaLife project [SeaLife] is a similar scenario, but in a different domain. Here, the users are Life Scientists, browsing web resources, for example the NELI site at http://www.neli.org.uk. Linking will be driven by existing vocabularies such as the Gene Ontology [GO]. An addition requirement for SeaLife is that different users require different behaviours. Thus a specialist may be interested in more detailed information than a casual browser. This introduces a potential need for application specific annotations to be associated with vocabularies.