Use of statistical techniques in Personal Retrieval Systems

Abstract

Personal Information Retrieval Systems (PIRs) for microcomputers have not yet had the same kind of impact as have Database Management Systems (DBMSs) for microcomputers. This may be attributed to the differences in the kinds of problems the two systems address. What a retrieval system can do, that a DBMS cannot, is to surmise what a record (document) is about. It can help find things that the user cannot clearly describe - sometimes even by subject matter. It is this capability that developers of PIRs should accentuate, because increasingly, personal databases are becoming sufficiently large so that finding relevant records is a recognized problem. If PIRs offer little more than DBMSs they will fail to meet that need. Fortunately, twenty-five years of research have produced some techniques (mostly statistical) that can improve the ability of retrieval systems to find relevant items. Even though these techniques have not been implemented by the large-scale retrieval services or mainframe-based retrieval systems, they could be; in fact, it is quite possible to implement them efficiently in a microcomputer-based PIR.