Solving problems in Library and Information Science using Fuzzy Set Theory

Library Trends, Wntr, 2002 by William W. Hood, Concepcion S. Wilson

As has been mentioned earlier, "relevance" is a concept that is not really dichotomous, and can readily be modeled by Fuzzy Set models instead. So Fuzzy IR systems work as follows: When documents are added to the system, index terms are assigned to the document, and each term is assigned a weight, indicating the degree to which that index term is associated with the document. The indexer is then free to indicate that a term applies only partially to a document, without having to make an absolute yes/no decision. Retrieval in a Fuzzy IR system is then based on Fuzzy Set algebra rather than Set algebra. The same Boolean operators are used (AND, OR and NOT), but the operators now rely on fuzzy union, fuzzy intersection, and fuzzy negation, rather than their classic (exact) equivalents.

This approach to IR has a lot of theoretical appeal, as it appears to be a much better model of the underlying process of selection (by users) of "relevant" documents. It is also a (relatively minor) modification of the traditional Boolean retrieval mechanism, so much of the existing infrastructure and mechanisms of IR are still valid. In addition, Fuzzy IR is more flexible in the assignment of index terms, with the use of partially relevant terms as well as fully relevant ones. The output can also be ranked according to relevance. Despite these advantages, there has not been much use made of Fuzzy IR in commercial systems. Reasons for this include: The cost of indexing continues to increase; many of the problems inherent in Boolean retrieval are still problems in Fuzzy IR; the capacity for ranking is not sensitive to all terms in the request; and traditional Boolean systems have done an adequate job in many situations (Bookstein, 1985, p. 124ff.).

Despite a lack of Fuzzy IR usage in most commercial IR systems, research has continued into the development of such systems, and there have been many applications in areas related to IR. Some of these will be listed below.

Expert systems and artificial intelligence. Gaines & Shaw (1985) discuss the history and development of expert systems, and the introduction of concepts from FST into this area. Graham (1991) also describes the use of fuzzy logic in commercial expert systems. FST has also been applied more generally in the area of artificial intelligence (Hofstadter, 1980; Winston, 1984). Nauck & Kruse (1999) use medical data to create fuzzy classification rules.

Knowledge-assisted document retrieval. A number of papers discuss the implementation of a knowledge-assisted document retrieval system (Subramanian, Biswas, & Bezdek, 1986; Biswas et al., 1987a, 1987b).

Relational databases with vague queries. Motro (1988) describes a database system that provides a user interface that permits vague queries based on FST. Some theoretical work done by Hashimoto (1985) can also be applied to Fuzzy databases.

Fuzzy clustering. The use of Fuzzy clustering algorithms in IR is also an area that has received a lot of attention. The excellent monograph by Miyamoto (1990a) provides a good description of the theory behind and the many uses of Fuzzy clustering. Fuzzy clustering can be applied in any situation where normal clustering is useful. Applications of Fuzzy clustering are described in Miyamoto, Miyake, & Nakayama (1983); Nomoto et al. (1987); Miyamoto, Midorikawa & Nakayama (1989); and Nomoto et al. (1990).