The field of Information Retrieval (IR) h as f ocused on the development and evaluation of retrieval models for text documents, such as those found in bibliographic databases (Rijs79, Salt83, Belk87]. Th ese retrieval models specify strategies for evaluating documents with respect to a given query, typically resulting in a ranked output. Hypertext researchers, on the other hand, have emphasized flexible organizations of multimedia% odes” through connections made with user-specified links, and interfaces that facilitate browsing in this network of links. A number of approaches to the integration of query-based retrieval strategies and browsing in hypertext networks have been proposed. The 13R system [Crof87, Thom89] and a medical handbook system described by Frisse [Fris88], for example, use query-based retrieval strategies to form a ranked list of candidate “starting points” for hypertext browsing. The 13R system also uses feedback during browsing to modify the initial query and locate additional starting points. The important issue from an IR perspective is the choice of a retrieval model, and consequently a retrieval strategy, for hypertext. This choice will have an impact on the effectiveness of retrieval and on the system implementation. A retrieval model can also provide a more formal specification of the meaning of some hypertext links.

It may appear that the highly connected structure of a hypertext database and the variety of link types that may be used to make those connections distinguishes it from a conventional text database. Network structures are, in fact, not new to IR and retrieval models have been proposed that use automatically and manually generated links between documents and the concepts or terms that are used to represent their content. Document clustering, for example, is a retrieval model based on links between documents that are automatically generated by comparing similarities of content [Wil188]. Citations are another form of link between documents that has been studied [Smal73]. Links between terms can be derived from term clustering [Spar70, Rijs77] or the information in a manually-generated thesaurus [Hump87]. Network representations that attempt to integrate this information have been developed [Oddy77, Crof87], but these have been used primarily for browsing. Recent studies have shown that retrieval effectiveness may be improved by combining the evidence represented by the different link types in a network representation