Home | Hypertext | IR Issues | Auto Link Generation | My System | Bibliography

Inference Networks

Indexes can also be considered as "precompiled links", providing immediate access to required information without navigating through the "document space" [Frisse & Cousins, 1989]. While nodes (containing information) can be considered part of the document space, indexes or index nodes can be treated as part of the "index space". In traditional full-text document retrieval systems, the document space and index space are essentially flat (they are together). Frisse and Cousins suggest the use of a hierarchical index space and a networked document space for information retrieval in hypertext systems. They have investigated the representation of index spaces as belief networks.

Croft and Turtle have also proposed an IR model for hypertext systems based on plausible or non-deductive inference using Bayesian inference networks [Croft & Turtle, 1989]. Belief networks or Bayesian inference networks are directed, acyclic dependency graphs where nodes represent propositional variables and links or edges represent probabilistic relationships between the propositions. A hypertext system can be compared to such a network - the roots of the dependency graph are hypertext nodes; interior nodes and leaves represent concepts.