Message extraction through estimated relevance

摘要

METER is a text analysis and retrieval system for non-expert computer users to exploit statistical associations between index terms of documents. It will run on a DEC PDP-11/45 minicomputer with continually changing collections of up to 20,000 documents at a time. A scaled version of METER with all major features of the full system has been implemented on a DEC PDP-11/70 as an experimental test bed for evaluation and comparison of associative retrieval algorithms. Although the basic structure of METER is similar to earlier statistical systems for retrospective document searches, the severe requirements of frequent updates of a document collection, of running on a small processor, and of meeting needs of users with little technical training have led to some novel developments. Among these are an update procedure that draws as much as possible on intermediate results from previous updates and a user interface that provides for control over the process of retrieval without calling for knowledge of how that process works.