The concept of forecasting based on semantic analysis of text information

Abstract

The main issue, which may be practically impossible to solve, is that in making a long-term forecast, all available data usually relate to the past or present, and there is not enough exact and reliable information for future issues. The following feature is observed: the larger the forecast horizon, the amount of information about the future state not only decreases, but also changes its structure. So, if all information will be presented as consistence of digital and text components, the relative amount of digital information for forecasting decreases, and the text information increases. In the limit, with a significant forecast horizon of decades, we will not have digital information at all, but only text information. The article considers the method of using text information to solve a wide range of problems in long-term forecasting. The method is based on K. Goedel’s idea of semantic truth in relation to the consistency of digital and textual statements. It is theoretically justified and practically shown that from these positions textual information after the implementation of the semantic analysis procedure, as a result of which when semantic cores of truth of statements are formed, they can be used for forecasting. This approach is methodically new and does not require the use of complex software tools. An algorithm for implementing the procedure for semantic analysis of information is proposed, as a result of which, a concept map is formed that allows analyzing the object of forecasting more systematically, without ignoring its elements and interconnections between them. Using the concept map, generation of new knowledge about the future States of the forecast object can be proceed and an “array” of possible options can be created. Subsequently, the most probable forecast data is selected from this “array”. The presented approach is based on using only text information without taking into account digital information, but in real forecast tasks, of course, digital information must also be taken into account. Obtained results illustrate the method and also interest wide class of long-term forecasting problems.