Computing with descriptive and veristic words: knowledge representation and approximate reasoning

Abstract

Knowledge representation is essentially composed of (1) rule pattern extraction and (2) selection of propositional formulas that represents the rule pattern extracted in propositional forms. Approximate reasoning on the other hand is also composed of (l) selection of propositional formulas and (2) choice of inference schemas such as interpolative reasoning or compositional rule of inference and/or combination of both. Rule patterns can either be extracted from experts via subjective interviews or from data via objective fuzzy clustering techniques. In either case, membership functions are determined for input and output variables and their linguistic terms. This forms a pattern to pattern representation structure of the rules that associate input patterns to output patterns. The rule pattern extraction is generally executed with unsupervised learning in case of fuzzy duster analysis techniques. Whereas propositional representation either in Type I or II formulas require supervised learning in order to determine the parameters of t-norms and t-conorms as well as the parameter of the convex linear combination of selected formulas and the parameter of defuzzification. Within the context of this framework of rule extraction and knowledge representation and approximate reasoning based on computing with words, we emphasize the interaction of descriptive and veristic words that generate Type II formulas and how these formulas could be applied for an improved reasoning schema.