Domain-Based Indexes: Indexing for Communities of Users

The formation of vocabulary evolves within communities, within domains of discourse. However, bibliographic databases have traditionally had one single index created for the entire database, even though bibliographic databases usually cover an arbitrary group of domains of discourse. As an experiment, indexes derived from the language used within one single specialized domain of discourse, a subset of a database. This radical departure from traditional practice shows significant improvements in retrieval performance. The conclusion that performance is best within specific domains and deteriorates as the scope of the system expands to include additional domains is consistent with experience in artificial intelligence and in machine translation. Analysis has required the development of an operational measure of the performance of intermediaries. Several theoretical and practical questions arise.