Relational information retrieval: Using relevance feedback and parallelism to improve accuracy and performance

This document describes how an information retrieval system may be treated as an application of a relational database management system (RDBMS) and thereby allow structured and unstructured data to be seamlessly integrated. It also demonstrates that an information retrieval system along with a variety of techniques for improving precision and recall of the search results can be implemented using standard SQL commands.; One of the ways precision and recall can be improved in an information retrieval system is through the use of relevance feedback. This document describes how different factors can influence relevance feedback and the individual contribution of each factor is explored and improvements over commonly used methods are demonstrated.; By using the relational mode as the basis of an information retrieval system, parallel processing is quickly, easily and transparently obtained when using a RDBMS implemented in a parallel environment. It demonstrates that in an information retrieval system running on a parallel RDBMS, processor CPU times and disk I/O's are less than 10 percent out of balance and it also demonstrates that an information retrieval system implemented in a parallel RDBMS can obtain significant benefits in terms of reduced processor CPU times and the number of disk I/O's when scaled up to additional processors.