IRF is a freely available object-oriented framework for information retrieval (IR) applications. A framework is software which defines the architecture of a group of related applications and supplies many of the basic components out of which they can be built, allowing application developers to extend the basic components to meet their special needs. IRF comes with a sample text IR application (SampleApp) built on the framework and two sample text collections (HCI and Cranfield).
We gratefully acknowledge that the starting point for IRF was the proprietary FIRE system developed in C++ by researchers at what was then UBS's Information Technology Laboratory. We appreciate their support. They deserve credit for many good design decisions that survive in IRF, which started as a translation of the core elements of FIRE. They are of course not responsible for any problems we may have have introduced into IRF in the translation and modifications we carried out.
Both IRF and the sample application are coded entirely in JavaTM. All the source code is freely available. All the classes but one were produced by the National Institute of Standards and Technology (NIST) in Gaithersburg, Maryland, USA. NIST is an agency of the U.S. government and by statute teh code we have produced is not subject to copyright in the United States. The BufferedRandomAccessFile class was developed outside of NIST and is covered by a GNU General Public License. Recipients of the software assume all responsibilities associated with its operation, modification and maintenance.
Please note that any mention of commercial products in the Guide or other IRF documentation is for information only; it does not imply recommendation or endorsement by NIST.
In early 1998 the NLPIR group at NIST began planning a follow-on to our public domain text information retrieval (IR) package ZPRISE - one which would be more portable, extendible, and better serve us and other research groups with increasing interest in non-text and mixed media information retrieval. IRF and applications built on it are intended to help meet the often stated needs of IR researchers, developers, and educators for software building blocks they can use as-is or open, study, and modify rather than reinvent. We decided that the best solution would be a freely available object-oriented IR framework.
We have accomplished much of what we set out to do:
We also investigated the extent to which we could stretch the current implementation to handle significant amounts of text e.g., a TREC-sized collection (500 mB - 2 gB). We did this as an experiment, knowing that the requirements for operational speed would often run counter to the main design goal of IRF. That goal is speed in (prototype) application code development due to:
Changes in the implementation might produce a faster application, but we are inclined to believe that the best use of IRF is in rapid prototyping of applications with small to moderate amounts of data to find the best approach. Once the approach has been determined, a largely separate real application or more nearly operational prototype can be built, sacrificing flexibility for optimal performance, as needed.
In making the code generally available, we invite others to use and improve IRF in whatever ways best serve their educational and research purposes.