There have been various attempts to automate the process of generating a hypertext. Node segmentation and link generation are two most difficult processes in an automatic hypertext generation system. Node segmentation separates a big node into several small ones which have the same title. The same titles of those nodes, which are resulted from node segmentation or homograph intrinsically, raise a new problem of link sense ambiguaties during the establishment of links among nodes. In addition, since the information of a real user's history path is not available until the time of browsing hypertexts, traditional automatic hypertext generation systems cannot use the information at all.
In this thesis, an automatic link generation model is proposed to resolve the link sense ambiguities. This model uses an approximation to the information of a user's history path, which is called the virtual history path information, thereby making it possible to reflect the information in time of creating hypertext. In the model is defined a sequence of node sets which roughly correspondes to that of nodes in the real user's history path. The sequence of node sets forms a virtual history path. Once a virtual history path has been established, it is used to compute score of each candidate node, and the node which gets best score is selected as link destination. The results from our experiments has revealed that the virtual history path information plays an important role in link sense disambiguation.