We believe to have demonstrated that the associative network idea for hypertext networks and the WWW can be taken further than the model. Hypertext systems can be equipped with Hebbian-style neural network learning rules and can, based on local browser navigation patterns, structure themselves from a completely random initial condition into meaningful networks. Resulting network structure reliably and validly resembles browser's associative intuitions and could enhance human browsing and retrieval efficiency. The system is furthermore simple: it does not require eloborate user-models (in fact, the network itself is the user-model), it does not require complex information be gathered and stored besides the actual hypertext systems. It is therefor fit to be used and implemented for the WWW, given a number of minor adjustments to the present HTTP protocol [Bollen&Heylighen, 1996]

Especially the WWW, due to its enormous size, could profit from a non-intrusive systems capable of automatically linking related information and restructuring existing but poorly organised parts of the network according to real user preferences. But, the learning algorithms can be expected to function just as well on stand-alone, moderately sized hypertext networks.

The system and its by-products (weigthed connections) can find many other applications. Weighted connections for example would enable new and flexible search mechanisms such as associatively browsing automatic agents and spreading activation. The implementation of automatic retrieval mechanisms that rely and partial and associative descriptions of the desired information such as spreading activation could greatly enhance the possibilities of WWW search engines.
An adapted set of learning rules could even introduce a new form of learning above the level of generating connections among related node: the automatic acretion of related nodes into by the WWW independently generated meta-nodes that represent original concepts and ideas.
The Adaptive Web Systems as proposed in this article does not just have its uses in the domain of hypertext and its related problems of navigation and network design. The large set of association norms that could be derived from its application to large hypertext networks such as the WWW, could by themselves be applied in many different domains. The automatic conversion of linear texts to hypertext could e.g. profite from accurate association norms for an as large number of words as can be found on the WWW.
Secondly, from a scientific point of view, association data from browsed hypertext networks could be used to compare the associative maps of different groups of people, different nationalities, ethnical and language groups, etc.

Apart from the practical implementation of this system for adaptive hypertext, further research will necessarily concentrate on establishing accurate models of human navigation behaviour. It needs to be determined wether human browsers effectively apply heuristic procedures to locate information in hypertext networks, and how or wether the relation between these browsing strategies and the required network structure can be quantitatively modelled and expressed as guide-lines to network designers.