6. Linking
There have been many attempts to classify links [Con87] [Fri90] [Myk95]. Designers of hypermedia documents need to decide about the main linking structure of their documents as well as about what additional linking facilities to supply to the user. To limit cognitive overload, it can be helpful to only provide a subset of different link types. We propose a link classification that closely fits the first 4 concepts for navigation in cyberspace:
- Page Links -- Sequentialization
- Page links rebuild the original ordering sequence of the nodes and thus reflect the primary sequential reading order as intended by the hypermedia author. They connect a particular page with its predecessor and successor and thus recreate a locally sequential context for the node.
- Hierarchical Links
- Hierarchical links reflect the logical ordering between the original sections of the document. As most of the documents are inherently hierarchically structured, hierarchical links are the most powerful means for achieving easily navigable and user friendly hyperdocuments.
- Similarity Links
- Similarity links connect nodes that have similar contents, but are no yet connected by page or hierarchical links. They are frequently computed automatically based on similar terms in different nodes and are generated by means of a statistical evaluation of the node collection. Similar nodes are ranked using the vector space model.
- Search Links
- Search links are generated automatically in answer to a query that was made by the retrieval system.
In addition to this first four categories of links that reflect the document structure and thus also offer a primary means for structuring information, we can identify another link category that embodies cross references between nodes:
- Reference Links
- Reference links are embedded into the original text and refer to other nodes or text passages within the original text, as, e.g., `see section 7', or [Knuth73]. Reference links are therefore a special case of similarity links.
This book describes many complex tools and techniques for navigation and information location in hypertext documents. In the best of all cyberworlds, there would be the ideal solution: If at any time the one link that the reader really wanted would be available, then there would be no need for sophisticated navigation tools like index, overview maps, fish eye views, cone trees, etc., described in later chapters. To satisfy this requirement, the hypertext system would have to figure out the current needs of the users based on their user model, browsing history, and current context, and then offer them the most meaningful link to the next node to be read. Unfortunately, the state of the art in automatic hypertext link generation will not be that advanced for a long time to come. Nevertheless, there are some promising approaches based on the concept of dynamic linking.