Information Retrieval Issues

While navigation or browsing is sufficient for small hypertext systems, more powerful information retrieval and indexing techniques are required for large scale hypertext databases.

• A browsing session can take a long time before converging to the required item or may not converge at all. Therefore, browsing mechanisms must be supplemented with querying techniques. In addition to content queries which retrieve the contents of nodes there should be structural queries to retrieve subgraphs of the hypertext network that match a given pattern. Query facilities which incorporate both content search and structure search can act as filters. Also, the implicit structure present in documents in terms of spatial characteristics such as geometry, distance, collocation, recurrence etc., can be analyzed to retrieve templates and document outlines.^{[Marshall & Shipman III, 1993]} Media-based navigation and picture-index techniques must be explored further to retrieve objects based on shape, color, motion, and auditory features.^{[Hirata et al., 1993]}
  
  
• Many researchers have investigated the possibilities of separating index information from contents thus forming an index space (or concept network) on top of a content space (or document network). These would not only facilitate IR but also accommodate dynamic linking and independent maintenance of the two networks. Further research is required in the area of aggregating hypertext networks into semantic or hierarchical clusters.

Existing query languages are suitable only for content-based searches.

• Query languages need to be extended to perform structural queries. These extensions include the notions of quantifiers, recursive operators, aggregation, and improved semantics. Visual query languages need to be explored further. Such visual queries should balance expressiveness with ease of use. Users should be able to retrieve information by specifying spatial properties, templates, shapes, color etc. They should also be able to express queries by selecting and manipulating visual representations of hypermedia objects. Such a query language is being developed as part of the Multimedia Object Retrieval Environment.^{[Lucarella et al., 1993]} It will be interesting to see the results of their work. More work is required in the use of belief networks or Bayesian inference networks for hypertext-based IR. The computational complexity of these approaches need further investigation.

Very little work has been done in the area of merging Artificial Intelligence with hypertext.

• A combination of inference-based IR and knowledge-based hypertext could greatly facilitate browsing and searching.

Formal methods and experiments are required to measure the effectiveness of these IR techniques.