Abstract

this report we describe the method and its application to two queries, both applied to a corpus of on-line Federal Register documents. In the first query we estimated the percentage of documents that were real and useful, rather than pseudo-documents related to the production usage of the Register. Our method allowed us to reduce the uncertainty of the estimate by a factor of 3:1. Our second query was based on a topic used in the nationwide TREC comparison of information retrieval systems. We found that there were probably several dozen relevant documents in the corpus that were undetected by any information retrieval system. This implied that the systems were not as accurate as they were reported to be. The report is intended to serve as a starting point for a user-friendly implementation of the system. To that end, it includes process schematics, dataflow diagrams, source code, and suggestions for GUI tools that could assist with the interactive aspects of the method.

Citations