The WWW consist of a huge collection of interlinked documents, each stored on a large number of individual servers. There are no instances above the level of an individual HTTP server. Any algorithm trying to improve/adjust network structure is therefore limited to what individual HTTP servers can do and measure.
The WWW is constructed so that the information available to individual HTTP servers is limited to where a certain browser came from and where he is going to. If it can be assumed that most browsers visit no more than 5 documents stored on the same server[Catledge&Pitkow, 1995], adaptive or dynamic hypertext systems will have to rely on very partial information concerning browser navigation behaviour. Than, on the other hand, the WWW's distributed nature also obstructs any attempts to control or adapt the networks structure beyond the scope of an individual server's content. The locus of control of any system for the automatic adaptation of structure for the WWW is limited to every small and local network stored on individual servers. Many of the existing systems for flexible hypertext depend on extensive information being stored and managed besides the actual hypertext network, such as users explicit appraisals of relevance [Mathe & Chen, 1992] or semantic knowledge networks that aid users in establishing useful connections [Berger et al., 1994], and they will therefore be very difficult or even impossible to implement.
In conclusion: any system capable of dynamically adapting network structure and content to the demands of its users must be able to deal with these limitations. It should: