AUTOMATIC DOCUMENTATION AND MATHEMATICAL LINGUISTICS最新文献

The author has presented possibilities for mathematical approximation and the dimension theory in computer science. The received formulas have led to a conclusion concerning existence of energetically-information equivalents thanks to which the theory of linearization of the nonlinear data and continuum digitization have received substantiation. The author has processed the obtained data from the perspective of the theory of dimension, and also has discussed hypotheses concerning an “error method” of chemical elements with negative mass.

Using the example of thermophysics, the evolution of approaches to working with scientific data on substances and materials properties is traced. This paper shows that across all stages, thermophysics can be classified as a data-intensive science, characterized by a focus on working with data, including its storage, its organization, and the extraction of meaningful information. It presents improvements in processing methods that are associated with the use of new information technologies, including machine learning techniques. Their potential for the field of thermodynamics relative to traditional statistical methods is analyzed. In this context, the general issue of the relationship between statistics and data science which has generated extensive debate in literature and online is discussed. All the authors’ conclusions are based on an analysis of specific issues related to the prediction of the properties of substances and constructing the equation of state and thermodynamic models for multicomponent systems.

The paper proposes an approach to the automated extraction and structuring of information from text, combining web scraping for data collection from online sources with a large language model for subsequent data mining. As a case study, texts from news publications on technology readiness levels from the CNews website were chosen to test the developed methodology in a specific domain. The model’s accuracy in identifying technology readiness assessments was 84–85%, which is comparable to similar results in other, less specialized tasks.

The subject domain, which can contain from one to several dozen or even hundreds (for large-scale systems) of subject lists, is a mandatory, core component of the vast majority of existing databases, around which all other components of the latter are built. The multilevel representation of objects in such databases is a standard practice aimed to simultaneously reflect both the numerous classifications that accompany subject lists and the key, basic relationship of materialization, which connects prototypes, models of subjects, and their instances. In spite of the existence of many methods for configuring subject schemes, the question of the invariance of such constructions is still relevant. In this paper, we show that the relationship of materialization is not the only constant in subject schemes. Multilevel models must include several standard relationships, such as relationships of variability and personification, which, together with the relationship of materialization, form an invariant framework of the subject scheme.

This article examines the concept of the rubricator (taxonomy) and its functions in media practice. The main shortcomings of content categorization in online media for mass audiences are identified. The role of the rubricator as a tool of information retrieval in the digital environment is emphasized.

Some possibilities of eliminating the consequences of the incompleteness of information in data analysis systems are considered, including through the exchange of tools between systems based on different theories.

The advantages and disadvantages of experts’ assessments of other experts’ competence in the task of selecting documents for library funds are analyzed, and it is proposed to use the methods of decision theory to improve the selection’s objectivity and quality. Statistical methods of assessing the experts’ competence used in libraries are described, allowing the assessment of the level of experts’ qualifications and the tracking of the dynamics of their competence in the process of conducting a collective examination by means of ILS. The importance of monitoring and adjusting the values of experts’ competence is shown to ensure a high level of quality in document selection, which will necessarily contribute to increasing the efficiency of library collections and the formation of high-quality collections that meet the needs of users.

Cryptography, which has a history going back many centuries, is currently experiencing a transition from the classical paradigm to a quantum one. Using bibliometrics, the dynamics of development and the ratio between three branches of cryptography—classical, postquantum, and quantum—in the period 1990–2020 are considered; the restructuring of the crypto-sphere in favor of the quantum paradigm since 2015 is noted. The world leaders in quantum cryptography research are industrialized countries, including the G7 group and China, and at the institutional level, leading universities, academic organizations, technology institutes, as well as large corporations, especially from Japan and the United States, and military research institutions, especially from the United States and China. The features of the quantum-cryptographic ecosystems of several countries are identified. It is shown that, in Russia, research is concentrated in the academic sector, universities and specialized organizations—the Academy of Cryptography of the Russian Federation, the Russian Quantum Center. By contrast with leading countries, the contribution of domestic corporations and small businesses is still limited. Geopolitical aspects of creating quantum-cryptographic security have been discussed. The Web of Science Core Collection database is used as a source for bibliometric analysis.

A study of the influence of differences in languages and training data on the quality of cross-lingual transfer of a trained speech model to Russian in the task of automatic recognition of emotions in speech is described. At the training stage, English, Polish, Chinese, and Japanese served as source languages, for which the IEMOCAP, nEMO, ESD, and JVNV emotional speech datasets were used, respectively, and the model itself was the HuBERT speech model on the transformer architecture. All models trained on the corresponding dataset were tested on a shortened sample from the Dusha Russian emotional speech dataset. Based on the data obtained, the main trends in choosing different languages for training the speech model and its subsequent transfer to Russian are considered, and differences in datasets are analyzed, which indicate the need for further work on collecting and labeling quality emotional speech data.

The paper proposes a new approach to automate processing of expert opinions in the information system in support of the competition of research works and projects based on the methods of decision theory using numerical and linguistic criteria. The methodology based on the integration of methods for assessing and comparing multicriteria alternatives and group methods of decision making includes the automation of the stages of experts assessing works and identifying the awardees and winners of the competition, taking into account the opinions of all experts. A specific example of the application of the developed methodology in practice is considered.