THE SITUATION OF UNCERTAINTY THAT ARISES IN THE PROBLEMS OF SEMANTICS AND WAYS TO SOLVE IT

Abstract

Various types of uncertainties that arise when solving semantics problems are considered. Decision theory investigates this situation involving incomplete input, current, and fuzzy information. But uncertainty in the problems of semantics has other manifestations. Its solution is carried out in different ways depending on its types. The problems of this class are related to recognition and when establishing the essence of certain objects, measures of similarity are introduced, which are a subjective assessment. For different measures, the values of the objective functions may differ due to the ambiguity of the result obtained for these functions or the chosen degree of similarity measures, and may not satisfy the purpose of the study. When choosing the result there is a situation of uncertainty. But with some measures of similarity, you can find a global solution. Such problems are divided into subclasses of solvable problems. Since the problems of semantics are reduced to combinatorial optimization problems, in which the argument of the objective function is combinatorial configurations, the situation of uncertainty may be related to the special structure of the set of combinatorial configurations. To solve it, it is necessary to enter several objective functions or to conduct optimization according to several criteria, which are reduced to a weighted criterion (linear convolution). Finding the optimal solution is carried out by self-tuning algorithms taking into account the constant and variable criteria, which are introduced in the process of solving the problem. That is, in the process of the algorithm generates additional current information (quality criteria), which affects the prediction of future results. The situation of uncertainty is manifested both due to developed fuzzy rules of information processing and evaluation and ambiguity in the choice of the optimal solution for several criteria in multicriteria optimization. To get out of this situation, self-tuning algorithms are developed, using the introduction of formal parameters in the process of solving the problem, which generates auxiliary current information that can not be specified in the input data. Also, subclasses of solvable problems are used to solve the situation of uncertainty, the reference library is structured to reduce unsolvable problems to solvable ones.