Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Maksim G. Shishaev has a total of 52 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Information Retrieval in Social Media Based on a Multi-Domain Dynamic Knowledge System

    by Vadim Pimeshkov, Marina Nikonorova, Maksim Shishaev

    The task of information retrieval is to find information that best satisfies the user's information needs. In the context of social media, information retrieval is complicated by the high dynamism of content, thematic heterogeneity, and the diversity of users' mental models. This paper proposes an approach to solving the problem of information retrieval under such conditions by constructing a multi-domain dynamic knowledge system. Its novelty lies in the combination of three levels of semantics: problem-oriented, represented by the ontology of the metatask (describing the search objectives); domain-specific, implemented through a dynamic multi-layer knowledge graph built on the basis of user content of social media; and domain-independent, based on a lexical database and a large language model. The knowledge graph allows us to reflect various contexts of concept usage corresponding to thematic clusters in the document collection. Such integration allows us to take into account the evolution of concepts, discourse features, and mental stereotypes of communication participants. To evaluate the effectiveness of the proposed system, an experiment was conducted using a dataset of publications from the VKontakte social network for problem-oriented monitoring of publications, where the selection of relevant publications from non-thematic sources is required. To solve this problem, a technology based on the use of the distance metric between query terms and publication terms in a multi-layer knowledge graph was proposed. The results of the experiment using this technology confirm the effectiveness of the proposed model for information retrieval tasks compared to standard keyword search and embedding models. In continuation of this study, it is planned to create a lexical database and also to consider the possibility of expanding the model by using a measure of pointwise mutual information and graph embedding methods.

  • TYPOLOGY OF AUTOMATED OPINION MINING PROBLEMS AND THEIR SPECIFICITY IN THE CONTEXT OF SOCIAL MEDIA ANALYSIS

    by Marina L. Nikonorova, Maxim G. Shishaev, Valerii A. Shtangel

    Interest in opinion mining and sentiment analysis of natural language texts is rapidly growing, despite the wide variety of problem formulations and ambiguity in the terminology used. This paper provides a review of the current state of the field of opinion and sentiment analysis based on scientific publications. The main types of opinion mining problems are systematized and a uniform formal description is provided. The specifics of solving opinion mining problems in social media and the main groups of methods applied to opinion mining are discussed. The results of this paper facilitate the interpretation of applied problems of natural language text mining, particularly social media content, into various types of opinion mining and sentiment analysis problems.

  • EXPRESS TECHNOLOGY FOR FORMING A TRAINING SET FOR PLANIMETRIC MINERALOGICAL ANALYSIS BASED ON MACHINE LEARNING METHODS

    by Vladimir V. Dikovitsky, Maxim G. Shishaev

    This paper examines the implementation of the planimetric method of mineralogical analysis as a classification problem. Machine learning is used to solve this problem, and an express technology based on data augmentation is proposed for generating a training set. A pre-trained neural network vectorizer based on the ResNet-18 convolutional artificial neural network is used to obtain the feature vector for classification objects. Experiments conducted using the analysis of apatite ores from the Khibiny deposit demonstrated that using this vector ensures a very high classification accuracy for pure apatite samples (up to 99 %, depending on the ore type and the size of the planimetric grid). Given the specific nature of the classification objects considered in this paper, a similar approach can also be used to determine the specific mineral content in ores using the areal method. The proposed express sample labeling technology ensures acceptable final accuracy for this problem with minimal effort required to generate the training set.

  • Open Access

    Machine learning-based mineralogical analysis using planimetric method

    by Владимир Витальевич Диковицкий, Максим Геннадьевич Шишаев

    Рассматривается проблема автоматизации минералогического анализа планиметрическим методом на примере апатитовых руд различных типов, характерных для Хибинского месторождения. Цель: изучение эффективности машинного обучения как средства формирования набора признаков и решения задачи классификации в различных постановках при планиметрическом анализе минералов. Результаты: Выявлены особенности планиметрического анализа как задачи классификации, в частности, определены свойства смежности и однородности классов в пространстве идентифицирующих признаков. Для различных вариантов формальной постановки задач классификации проанализированы потенциальные систематические ошибки определения содержания полезного компонента планиметрическим методом. Экспериментально подтверждена возможность непосредственного использования предобученной сверточной сети ResNet18, без изменений архитектуры и дообучения, для формирования признакового вектора объектов классификации, обеспечивающего хорошую разделимость классов. На примере рассматриваемых руд экспериментально подтверждена высокая эффективность (более 98% точности) применения нейросетевого классификатора и векторизатора ResNet18 для идентификации элементов изображения, относящихся к чистым классам «апатит»/ «не-апатит». Высокая точность классификации сохраняется при уменьшении размера ячейки планиметрической сетки вплоть до 2 × 2 пикселей (78%), а при размере ячейки 20 × 20 пикселей приближается к 100%. Исследована эффективность применения нейросетевого подхода к задаче определения удельного содержания полезного компонента в руде. Эксперименты не подтвердили эффективность реализации планиметрического анализа как задачи мягкой классификации без существенных модификаций архитектуры нейросетевого классификатора, однако показали высокую эффективность подхода при мультиклассовой постановке задачи. Абсолютная ошибка в точности определения содержания полезного компонента в последнем случае зависит от количества классов и типа руды и в худшем случае не превышает 6%, что выше точности экспертных оценок опытными рудничными геологами. Практическая значимость: подход применим для создания недорогих, быстродействующих и эффективных экспресс-анализаторов руд, не требующих специализированного оборудования.

  • FORMATION OF MULTILAYER KNOWLEDGE GRAPHS BASED ON THEMATIC TEXTS MODELING

    by Vadim Pimeshkov, Marina Nikonorova, Maxim Shishaev, Ivan Vishnyakov

    The paper considers the problem of forming multilayer knowledge graphs for use in intelligent analysis of social media data. The technology is based on the «soft» structuring of the existing document pool in accordance with the constructed topic model. Experimental testing of the technology and comparison of the formal properties of knowledge graph models obtained using topic modeling and the traditional method were conducted. The results of the comparison indicate the potential effectiveness of using multilayer knowledge graphs built on the basis of a topic model of texts when implementing intelligent data processing procedures in the conditions of multiple and dynamic subject areas.

  • THE POTENTIAL OF USING GENERATIVE ARTIFICIAL INTELLIGENCE TO SIMULATE THE RESILIENCE OF REGIONAL SOCIO-ECONOMIC SYSTEMS

    by Vitaliy Bystrov, Darya Khaliullina, Maxim Shishaev

    Minimization of human involvement in the construction of simulation models is an urgent task, the solution of which is in demand in the practice of modeling the life cycle of complex dynamic systems. One of the ways to solve it is to adapt existing solutions in the field of generative artificial intelligence to the specifics of developing computer models of complex systems. The article discusses the main areas of automation of simulation modeling using modern artificial intelligence technologies. The authors propose concepts for using large-scale language models to generate scenarios for simulation modeling of the resilience of regional socio-economic systems. Theoretical and practical issues of implementing the proposed solutions in modern simulation environments are considered. A description of a generalized algorithm for assessing the resilience of regional systems "Business-Community-Government" is given, built on the basis of integrating simulation modeling and generative artificial intelligence.

  • Open Access

    Гибридная технология синтеза транспортно-логистических систем на основе машинного обучения и имитационного моделирования

    by Елена Олеговна Неупокоева, Виталий Викторович Быстров, Максим Геннадьевич Шишаев

    В настоящее время наблюдается мировой тренд в использовании интеллектуальных технологий в комбинации с компьютерным моделированием для решения сложных прикладных задач. Текущая работа посвящена созданию программных инструментов для информационно-аналитической поддержки управления транспортно-логистическими системами. В статье рассматривается гибридная технологии синтеза транспортно-логистических систем на основе машинного обучения и имитационного моделирования, позволяющая формировать и анализировать конфигурации сложных транспортно-логистических систем с большим количеством компонентов. Приводится краткий обзор нейросетевых методов и технологий, применяемых для решения прикладных задач оперативного управления и планирования в транспортной логистике. Авторы предлагают формализацию задачи планирования конфигурации транспортно-логистической системы в общей постановке и в частном случае. Рассматривается разработанная компьютерная модель имитации сценариев реализации плана движения транспортных средств на заданной конфигурации транспортно-логистической сети.


  • Open Access

    A Combined Term Extraction Method for the Problem of Monitoring Thematic Discussions in Social Media

    by Вадим Константинович Пимешков, Марина Леонидовна Никонорова, Максим Геннадьевич Шишаев

    Term extraction is an important stage in the automated construction of knowledge systems based on natural language texts, since it provides the formation of a basic concept system, which is then used in applied problems of intellectual information processing. The article discusses the problem of automated extraction of terms from natural language texts for their further use in the construction of formalized knowledge systems (ontologies, thesauruses, knowledge graphs) within the problem of monitoring thematic discussions in social media. This problem is characterized by the need to include in the formed knowledge system both concepts from several different domains, and some general concepts used by the audience of social media within thematic discussions. In addition, the generated knowledge system is dynamic both in terms of the composition of the domains it covers and the composition of relevant concepts to be included in the system. The use of existing classical methods for term extraction in this case is difficult, since they are focused on extracting terms within one domain. Based on this, to solve the problem under consideration, a combined method is proposed, combining approaches based on dictionaries, NER tools and rules. The results of the experiments demonstrate the effectiveness of the proposed combination of approaches to term extraction, which makes it possible to extract terms for the problem of monitoring and analyzing thematic discussions in social media. The developed method significantly exceeds the precision of the considered term extraction tools. As a further direction of research, the possibility of developing a method for solving the problem of identifying nested terms or entities is considered.