AUTOMATIC DOCUMENTATION AND MATHEMATICAL LINGUISTICS最新文献

The article examines the evolution and current state of the data intensive sciences (DISs). The article focuses on approaches to methods of data mining generated by the development of artificial intelligence. It is noted that the rich opportunities of new approaches have caused unreasonable enthusiasm among scientists with respect to their capabilities, while the achieved level of knowledge is clearly ignored. It is shown how numerous facts of limited data processing potential have gradually accumulated without taking into account all previously established laws of nature and research methods. A significant role in the awareness of the real potential of working with data (including big data methods) was played by specialists in the field of methodology of science, who created a new direction, the epistemology of the DIS. Various ways and means of introducing expert knowledge at subsequent stages of analysis in the form of machine learning are listed. In sum, the appearance is noted of special algorithms for physically informed machine learning using data in combination with a traditional approach based on solving equations of mathematical physics.

The experience and specifics of the use of intelligent data analysis (IDA) in high-tech medical diagnostics are discussed. The current version of the IDA is a mathematical formalization of the so-called causal similarity heuristic by algebraic means. The main features and abilities of the developed approach are demonstrated in relation to the tasks of the diagnosis and treatment of certain types of human brain tumors. Some results characterizing the causality of the effect of pseudo-progression and tumor recurrence are presented. The potential and prospects of the developed approaches and diagnostic tools in the arsenal of modern evidence-based medicine are considered.

Three main stages in the approach to identifying news on natural disasters and the clustering groups of citizens are considered. The first step presents the sequence of performing several natural language processing tasks. The problem of the ambiguity and vagueness of news similarities has been ignored in traditional methods of event detection. To this end, the second step is to apply fuzzy set techniques to the events extracted to improve the quality of clustering and to eliminate the vagueness of the extracted information. A certain degree of hazard is then entered as input to the citizen clustering method to identify communities that feature similar degrees of distress. The results show that with the help of the proposed approach, it is possible to identify homogeneous and compact clusters of citizens.

A cluster analysis of the 100 most cited scientific papers on bibliometrics (scientometrics and informetrics) is carried out. The main bibliometric methods are identified, which make it possible to distinguish between the subject areas of their application. Co-citation of scientific papers and co-citation of their authors are still common. The analysis of scientific vocabulary (co-occurrence of words) complements the noted techniques. More often, in recent years, a set of bibliometric methods is used as a result of the application of modern computer programs to the statistical processing of bibliographic data. The fields of development and the application of bibliometrics related to separate clusters include nonmatching groups of various research topics in combination with different bibliometric methods. The use of bibliometrics in soft science (such as the social sciences) research is widespread, and in recent years it has not come at an inferior rate to its more traditional use in the natural sciences. The statistics on documents, references, and other indicators, to some extent, form the soft sciences themselves and delineate the areas of relevant studies.

This article discusses the importance of the timely receipt and analysis of information on the likelihood of competitors producing similar products, especially for military and dual-use products. Methods of information collection are considered. The main form of information collection is monitoring. By processing information about direct and indirect competitors, both open and closed, specialists in Russian enterprises and companies, knowing the weaknesses and strengths of their competitors, change their production process over time. The main characteristic of the information system for obtaining and processing data is bandwidth, as the system that can provide more bandwidth wins the competition. That is, one criterion for the competitiveness of any system is the bandwidth of its information channel. The process of forming the levels of hierarchy of the information system is considered. The importance of information systems in the modern economy is noted, as in a business where new competitors are constantly emerging, maintaining one’s market position requires constant attention.

In this paper, we propose techniques to improve the efficiency of systems acting with knowledge bases and based on functional neural networks (FN-networks) formalism due to the use of their structural features. The article addresses the case of partially defined FN-networks, which have a regular structure and are set as lists of multiple similar objects and fragments with common structure such that their number is unknown and may be unlimited. It offers an algorithm to find solution, which is based on dynamic formation of limited, fully determined local fragments of partially defined network and subsequent transfer of the results to the entire FN-network, which may be endless.

To unify the understanding of and improvements to management efficiency, the conceptual apparatus of information support for risk management of BP is clarified, and methods of analysis and assessment of risk in the information environment are explored, using the example of a metallurgical company. The development of economic and mathematical support in calculating a quantitative measure of risk is presented, and a matrix for calculating the correction weight coefficient W_i for the determination if this measure is proposed, which allows us to take into account the duration and importance of identified risk factors, rank risks, and make informed decisions based on a more complete and detailed assessment of them.

The paper proposes the use of a probabilistic knowledge-extraction mechanism to resume the use of good old-fashioned artificial intelligence. As a model task, it is proposed to automatically generate the rules of the Tic-Tac-Toe game. Similar rules were used in the well-known General Problem Solver system developed by A. Newell and H. Simon. Unlike the past, where scientists wrote the rules for the optimal game strategy manually, we will use a probabilistic-combinatorial formal method to generate them automatically. We will also discuss the relationship between the proposed approach and the reinforcement-learning paradigm.

An approach to implementing perceptual-cognitive interfaces (PCIs) is presented in the context of the synchronous transmission of speech messages with perceptual semantics via technologically available and relevant sensory channels (visual, auditory, tactile, olfactory, and motor). The architecture of interfaces, implementation of the language model, and limitations imposed by the PCI model are considered. The introduction of the term “sensory technolinguistics” is justified, and possible areas of using perceptual-cognitive interfaces are described.