Author/Editor     Vizjak-Pavšič, M; Musek, J; Rajkovič, V
Title     Razumljivost baz znanja kot dejavnik učinkovitosti ekspertnih sistemov za podporo odločanju
Translated title     The comprehension of knowledge bases as a factor of effectiveness of expert systems for decision support
Type     članek
Source     Anthropos
Vol. and No.     Letnik 27, št. 5-6
Publication year     1995
Volume     str. 47-67
Language     slo
Abstract     The present study reports on the results of the experiment to test the comprehensibility of four knowledge representation modes in the expert system shell DEX (tables of aggregate rules, decision trees, three-dimensional graphs, regression weights of individual criteria) and explains the sources of variability of the results. The shell of the DEX expert system, developed in the Artificial Intelligence Laboratory at the Jožef Stefan Institute in Ljubljana, is based on the methods of cybernetics and artificial intelligence and is intended for the search for correct decisions, above all in solving complex, multiparameter problems. As shown by the results of multivariate analyses of the collected data there are significant differences in comprehensibility among the studied knowledge representation modes of the DEX expert system. Decision trees, followed by aggregate rules have the highest comprehension indices (Ic), while three-dimensional graphs and regression weights of individual criteria achieve consinderably lower Ic values. An advantage of decision trees over other studied knowledge representation modes is in their structure, which enables quick recognition of the inherent hierarchies and direct, cognitively economical study, from the root of the decision tree to the corresponding leaf, whereby it is not necessary to review the complete presentation of data, which is especially important in more complex representations. As the results of this showed, decision trees are the only representation mode studied in which comprehensibility does not decrease in conditions of higher levels of complexity. Their comprehensibility also varies very little with respect to the semantic content of concepts or the type of information contained. A distinctive advantage of decision trees over other presentation modes of logical rules of the DEX expert system was also demonstrated by scores on 7-point scales for the evaluation of their comprehensibility, clarity and difficulty.
Descriptors     ARTIFICIAL INTELLIGENCE
EXPERT SYSTEMS
DECISION MAKING, COMPUTER-ASSISTED
DECISION SUPPORT TECHNIQUES