Intelligent Information Approaches

Many have attempted to make information systems more intelligent. This is usually done with expert systems, knowledge based approaches, modeling of human search intermediaries, special inference techniques, or learning approaches.

• Expert Systems --- Often for a Library Reference Desk --- to help automate the activities of reference librarians and the reference area of libraries, by encoding the knowledge of these expert intermediaries.
• Concept-Based Retrieval --- to allow people to ask questions and find items on the right concept, thus going beyond keyword systems. This involves various ways to describe concepts and to combine them to describe information needs. One commercially available system with this orientation is TOPIC.
• NLU-Based Retrieval --- to allow computers to better retrieve relevant documents, and to understand natural language questions, natural language understanding methods are needed. Many of these deal with such problems as:
  • identification of phrases
  • identifying names, dates, times, locations, abbreviations
  • finding related terms to those in a query, for query expansion
  • recording desired relationships between concepts

  The Department of Defense has sponsored the MUC (Message Understanding Conference) and TIPSTER/TREC (for extraction of facts from texts, and for retrieval from large text collections) projects to encourage competition in these areas.

• DEBIS --- Distributed Expert-Based Information Systems: an approach to building intelligent information systems as a collection of communicating modules, each incorporating knowledge-based or expert systems techniques. Often this involves a module for each of the functions carried out by human intermediaries, like problem description or response generation. For a detailed discussion see article by Belkin et al. in Information Processing and Management, Nov. 1987. Typically there are: multiple and/or distributed data, information, knowledge bases. As in expert systems, these systems attempt to record expertise:
  • of human (search) intermediaries
  • of information access methods
  • of subject domain

  Prototype systems include U. Mass. Amherst's I3R system and Virginia Tech's CODER system

• Probabilistic Networks --- an approach that builds upon Bayesian inferencing methods. An example of this is U. Mass. Amherst's Inquery system
• Learning --- to adapt to a user, possibly using feedback information. Two prominent methods use:
  • Neural Networks --- that is trained from various trials, perhaps to see how important various keywords or combinations are in connection with a particular person or information need.
  • Genetic Algorithms --- for example, that can find a better query, given information on various other queries and retrieved documents.