Інформаційна система виявлення прихованого змісту текстових повідомлень. Частина 2. Розробка аналітико-математичного забезпечення інформаційної системи виявлення прихованого змісту в текстових повідомленнях

Abstract

Current work consists of two chapters. As about first section, it can be concluded that the use of classical supervisory machine learning algorithms is often not suitable for applications related to text analysis because of the complexity of the markup. Algorithms of learning "without a teacher", in turn, for most applications give poor results. A promising area of research is the inductive construction of the rules of CPSL (for extracting information) or another language with no less expressive means (for combining results). It should pay attention to the following aspects: – Use of active learning. – Increased interactivity. – Use of hybrid teaching methods. – Bootstrapping is an approach based on the fact that the results obtained at a certain iteration of training are used to prepare the input data of the next iteration. At the same time, one can use achievements from the field of inductive logic programming, as well as combining the descending and ascending approaches, when two boundaries are considered for each hypothesis are the most general and the most specialized, taking into account the examples viewed. In the second section, a mathematical model of learning the developed system was developed. In the third part, a set of UML diagrams was developed, which fully describes the classifier developed in the fourth part. In the fourth part a classifier was developed that could recognize text from any image. The classifier is developed in Matlab 2016b environment. This classifier is needed in order to prevent anything from getting a text message to further distinguish it from a hidden context.