AN EXPLANATION MODEL IN AN INTELLIGENT SYSTEM AT THE LOCAL, GROUP AND GLOBAL LEVELS OF DETAIL

Abstract

The subject of research is the process of formation of explanations in intellectual information systems. Machine learning methods are used in modern intelligent systems. The process of obtaining the solution formed on the basis of such methods is usually opaque to the user. As a result of such opacity, the user may not trust the solutions proposed by the intelligent system. This reduces the efficiency of its use. Explanations are used to increase the transparency of decisions. The explanation is represented by knowledge about the reasons for the formation of the result in the intellectual system, as well as about the reasons for individual actions in the process of formation of the result. Also, the explanation may contain knowledge about the influence of individual functions on the results obtained by the intelligent system. Therefore, it is advisable to form an explanation at different levels of detail in order to show both the generalized reasons and effects on the obtained decision, as well as the reasons for choosing individual intermediate actions. The purpose of the work is to develop a generalized model of explanation considering the states of the decision-making process in an intelligent system to build explanations based on known data regarding the sequence of states and the properties of these states. To achieve the goal, the following tasks are solved: structuring the properties of explanations; determining the possibilities of approaches to building explanations based on the states and structure of the decision-making process, as well as on the basis of input data; construction of an explanatory model. Conclusions. A generalized model of explanation in an intelligent system for local, group and global levels of detail of the decision-making process is proposed. The model is represented by an ordered sequence of weighted dependencies between events or states of the decision-making process. The model is focused on presenting the possibility to highlight a local explanation within the framework of a global explanation and to present a chain of group explanations between the events of obtaining input data and the resulting decision. In practical terms, the proposed model is intended for the construction of explanations using approaches based on the simplification of the process of functioning of the intelligent system and on the basis of highlighting the influence of individual functions and actions on the final result. Additional capabilities of the model are related to the detailing of the events of the decision-making process from the selection of individual variables that characterize the state of this process, which makes it possible to form an explanation based on the use of known concepts and concepts in the subject area.