A COMPOSITE MEASURE FOR WEIGHTING DATABASES IN DEFENSE, ENGINEERING, AND SCIENCE (MULTIDATABASE SEARCHING, ONLINE SEARCHING, DIALINDEX).

摘要

The primary problem of this dissertation is to propose a composite measure as a technique for measuring the relevancy of databases. The databases are characterized as single units by the measure of closeness, C(,M), values. The measure of closeness consists of two weighted factors: (1) a relevance factor, and (2) a descriptive factor. The relevance factor is the sum of the recall and precision ratios. The descriptive factor is the sum of the weighted properties of each file as follows: (1) subject coverage, (2) thesaurus strength, (3) technical level, (4) subject coding, and (5) length of years searched retrospectively.;For each of the seven topics, online searches were performed on a group of databases. These databases, ranked according to C(,M) values, were compared with their corresponding databases ranked by retrievals from DIALINDEX, a Dialog multidatabase file. The first experiment compared six randomly selected Dialog files and Dialog files subjectively selected for their expected higher relevance to the topics. While randomly selected files retrieved some relevant citations, these files generally did not contain many relevant citations. The second experiment compared the DIALINDEX method and the measure of closeness, C(,M), technique.;Mann-Whitney two rank and Spearman Rho rank correlation tests failed to indicate conclusively that the DIALINDEX method is different from use of the weighted measure of closeness alone. The tests did indicate DIALINDEX term frequency retrievals appear to result in ranking relevant databases. Possible artificial intelligence designs may further enhance the future modelling of weighting schemes for more effective multivendor and multidatabase online search techniques.;Only unclassified terms, titles and/or abstracts were discussed in order to conform to U.S. national security requirements.;Two experiments were conducted to test if the measure of closeness may be utilized to select the relevant databases in DIALINDEX searches in the general areas of defense, engineering, and science. Databases from Dialog Information Services, Inc., Defense Logistics Studies Information Exchange, Defense Technical Information Center, Mead Data Central Nexis, NASA/RECON, and DOE/RECON were also used. Searches were conducted in seven sample topics: (1) composites, (2) missiles, (3) rockets, (4) sonar, (5) torpedoes, (6) underwater acoustics, and (7) underwater weapons.