KEYNET: An architecture and protocol for
high-performance semantically rich information retrieval
J. Elliott Smith
College of Computer Science
Boston, Massachusetts 02115
FAX: (617) 373-5121
We propose an architecture and protocol for semantically
rich information retrieval from a subject-specific corpus of
The technique assumes that information objects have been labeled
using small directed graphs, called content labels.
A content label subsumes the role of an abstract.
Both content labels and queries are required to conform with a general
framework (ontology) for the subject area.
Our method avoids explicit isomorphism testing by decomposing queries
into fragments of small maximum size and using tables of these fragments
to accomplish comparisons.
The index structure is compatible with distributed processing architectures
and scales up well, allowing very large collections to be searched
very quickly using semantically complex queries.