A model for a weighted retrieval system

Abstract

There has been a good deal of work on information retrieval systems that have continuous weights assigned to the index terms that describe the records in the database, and1 or to the query terms that describe the user queries. Recent articles have analyzed retrieval systems with continuous weights of either type andlor with a Boolean structure for the queries. They have also suggested criteria which such systems ought to satisfy and record evaluation mechanisms which partially satisfy these criteria. We offer a more careful analysis, based on a generalization of the discrete weights. We also look at the weights from an entirely different approach involving thresholds, and we generate an improved evaluation mechanism which seems to fulfill a larger subset of the desired criteria than previous mechanisms. This new mechanism allows the user to attach a "threshold" to the query term. Statement of the Problem A. Boolean and Fuzzy SystemsA set of documents D* exists for retrieval. A set of index terms I* also exists. In a traditional retrieval system, a function F exists mapping D* X I* into [0,1] . For d ED* and T E I*,A Boolean query Q can now be constructed from the terms in I* and the operations AND, OR, and NOT. Given a query Q , a retrieval status value (RSV) can be obtained for each document d by using the usual truth table methods for the operators. The RSV is 1 (true) if d is "about" Q and 0 (false) if d is not "about" Q. A strictly Boolean retrieval system will retrieve all the documents for which the RSV is Looked at in another way, a term T defines a set T*, the 1 PI. set of documents "about" T :

Fuzzy versus probabilistic models for user relevance judgments

“…Both the probabilistic (Bookstein, 1985;Robertson, van Rijsbergen, & Porter, 1981;van Rijsbergen, 1979) and the fuzzy (Bookstein, 1985;1980;Buell & Kraft, 1981;Waller & Kraft, 1979) models have been examined within the framework of IR systems and applied to the design of indexing methods and matching functions. Both models begin by assigning numerical weights to individual terms in documents (and possibly in queries).…”

“…However, queries or topics may be represented by Boolean combinations of terms. The model may be used to determine membership of a document to the set of documents about such complex topics (Bookstein, 1980;Buell & Kraft, 1981). The procedure utilizes the membership values of the document to sets corresponding to the subtopics of the query.…”

“…However, the articles by Radecki (1982) and Croft (1986) show how Boolean logic may be included while computing document ranks. For the fuzzy model there are articles (for example Bookstein, 1980;Buell & Kraft, 1981) on ranking documents against Boolean queries. In general, the probabilistic model is based on an elegant theory, while the fuzzy model process has faced some criticism.…”

An inferential approach to Information Retrieval and its implementation using a manual thesaurus

“…One alternative is to evaluate P(d -d') as a fuzzy degree of relatedness between the sets d and d'. This is the method used in the previous approaches based on fuzzy sets (Buell and Kraft 1981;Kraft and Buell 1983;Radecki 1979;Waller and Kraft 1979). Yet another solution, which leads to a higher inference power, is to consider a document description as corresponding to a fuzzy world in fuzzy modal logic (Schotch 1975;Ying 1988).…”

Inductive Information Retrieval Using Parallel Distributed Computation.

“…There are a variety of methods in the information retrieval literature for assigning a relevance measure to a document (Bfinschi & Frei, 1982;Bookstein, 1980;Buell & Kraft, 1981;Salton, Fox, & Wu, 1983). Some use fuzzy logics to evaluate Boolean queries; others, known as vector models, treat documents and queries as vectors in a concept space of large dimensionality and then compute the distance between vectors.…”