next previous contents
Next: Case study: KBAS-TW Up: The KMS scenario Previous: Application and measurement criteria

Summary

Current IR approaches are based on character string oriented techniques and inverted files. At present, there is no full integration of computational linguistic techniques, apart from morphological approaches or shallow syntactic analyses (cf.below). Their typical functionality accounts for:

The obvious advantages are application independence and near language independence (so-called cross-products). However, precision, recall and specificity have to be improved. The limitations of systems are mainly based on the expressivity of the information language (cf.ãbove).

Current trends in IR can be characterised by the techniques used:

  1. Statistics-based;
  2. Menu- or dialogue-based with relevance feedback and user models;
  3. Natural language based, e.g. thesauri, synonym techniques and generation of inflected forms;
  4. Knowledge-based.

The first three techniques offer well-established formalisms which allow for rich expressiveness and reliable deduction mechanisms. However, regarding the fourth technique, at present knowledge representation is associated with individual system design. This is because procedural construction of a knowledge base (KB) is not permitted. Thus, one finds that idiosyncratic interpretations of the predicates and other constructs of the KB naturally occur. In addition, KBs have a high amount of documentation effort and are limited to small worlds. This latter limitation is a crucial restriction on the extensibility of a KB as well as on the possibility of changing application domain. In the following section, we will present potential solutions to solve some of these shortcomings.



ceditor@tnos.ilc.pi.cnr.it