![]() |
![]() |
![]() |
![]() ![]()
Sign Up Now
Sections
Support
Featured Sites
![]() ![]() |
![]() |
Search Home | Advanced Search | Search Help | Search the Web |
![]() March 16, 1998, TechWeb News Fuzzy applet performs smart database search By R. Colin Johnson
Wayne, Pa. - A new approach to database manipulation employs fuzzy-logic concepts to make "intelligent" choices from imprecise and perhaps conflicting queries. The significance of the fuzzy formulation lies in its flexible ability to get by on whatever level of knowledge a user is able to impart. Pei Wang, a former Indiana University researcher who is now a software engineer at Explore Reasoning Systems Inc., here, has posted his system to the Internet as a freeware Java applet. Internet surfers can experiment with various kinds of fuzzy ranking, picking the top few candidate rows from any tabular data according to user-specified requirements. For example, in data provided to illustrate the system, users specify computer-hardware requirements. The fuzzy-recommendation system then picks the top 10 candidates from a database of computer systems. Any tabular database, however, can be pasted into the Java applet to replace the computer-hardware database. A fuzzy-recommendation system differs from a database-query system in that the imprecise and conflicting requirements can be resolved to hone the resulting candidate list. Conventional database-query languages can barely handle ranges, much less imprecise specifications and conflicting requirements, although such problems abound in real-world data use. As a result, a user's requests often cannot be represented as proper database queries. For instance, in the example system, a user can ask for a "very fast" computer rather than specify a clock speed like 100 MHz. But fuzzy systems really shine when requirements are conflicting, such as when a prospective buyer wants a "very fast" computer that is also "low cost." Each time the buyer specifies a new requirement, the problem's complexity increases geometrically for conventional database search engines, producing too many useless results. But with fuzzy logic only a few rules need to be added. The inference and defuzzification methods resolve conflicts among all the rules no matter how many. The result is a ranked list of "best" candidates . "What really matters in fuzzy recommendation systems is the relative value of items, not the absolute value of any specific attribute. For instance, 11 a.m. may be considered very close to 9 a.m. if all other flights depart in the afternoon," Wang said. In operation, each attribute of any item in the Java database-the columns of the spreadsheet-like display-are processed separately to find the optimal item for that dimension of the problem. After assigning a score to the best candidate for each attribute, the system infers the weighted sum of all the optimal candidates. The best overall choice, thus calculated, reflects an "intelligent" choice. "In this way, conflict requirements can reach a reasonable overall compromise, such as when a computer is just a little slower but much, much cheaper," said Wang. Using SmartRanker consists of pasting in spreadsheet-like tabular data into the "input" area of the Java applet. As many as 200 rows with 10 columns can be processed in the small applet. SmartRanker selects first a fuzzy "filter" for each data column, then a fuzzy "ranker" aspect of the requirements for each column. Users can also specify weights for the importance of each column, preferring, say, low price over high speed, or vice versa. SmartRanker is automatically loaded and available for data sorting with a Java-enabled browser found at www.cogsci.indiana.edu/farg/peiwang/SmartRanker/SmartRanker.html and is based on a paper by Wang: ftp://ftp.cogsci.indiana.edu/pub/wang.fuzziness.ps. Copyright (c) 1998 CMP Media Inc.
http://www.techweb.com/se/directlink.cgi?EET19980316S0048 |
![]() |
![]() |
|