Summary and Future Work

An IR model has been introduced that generalizes many commonly used IR mechanisms. A distributed architecture and indexing algorithm have been developed for this model. A prototype system has been developed to measure the performance characteristics of the algorithm, and the prototype has achieved a throughput of 500 queries per second. The model can support semantically complex content labels and queries.

A number of research efforts are actively being pursued on both KEYNET and closely related projects.

• Since a KEYNET ontology is required for indexing a corpus using a KEYNET search engine, it is important to have good tools for building such an ontology. The ontology is also closely related to the user interface to the KEYNET system, making the ontology even more fundamental. The OntologyBuilder[BF93] is a research effort related to KEYNET that addresses the problem of building and managing ontologies. Even a small ontology can occupy hundreds of megabytes of storage, so database management is clearly an important part of this problem. The OntologyBuilder uses object-oriented database management tools for managing the very large and complex data structures of an ontology.
• The indexing strategy used by KEYNET is a distributed hashing method. One feature it does not yet have is expandability. Several distributed, expandable hashing methods have been proposed, and we plan to integrate KEYNET with one of them.