Yuriy Evegenievich Gapanyuk

Web tables are an important source of structured data for question-answering systems; however, retrieval methods differ substantially in their retrieval units, structure modeling strategies, and mechanisms for matching queries with table representations. The purpose of this article is to systematize methodological directions in structure-aware retrieval of web table fragments and to identify limitations that motivate the use of multivector and late-interaction retrieval schemes. The study is conducted as a scoping review. The search was performed in Scopus, OpenAlex, eLIBRARY/RSCI, and through citation analysis; from 767 identified records, 42 studies were included in the final synthesis after duplicate removal, screening, and full-text assessment. The article proposes a unified formal framework for comparing retrieval methods, develops a taxonomy of approaches, and analyzes retrieval granularity, vector interaction types, relevance aggregation mechanisms, structural signals, and computational limitations. The results show that dense and structure-aware whole-table methods dominate, whereas scalable retrieval of rows, cells, subtables, and multi-table evidence remains insufficiently developed. The article concludes that further research should focus on multivector methods that preserve local table structure while maintaining acceptable computational complexity.

This paper addresses the operation of a metagraph transformation algorithm. Graph-based models, including complex graph architectures, are being extensively utilized for characterizing the structures of complex systems. The significance of this study resides in its analysis of a transformation algorithm for large-scale complex systems represented as metagraphs. It discusses the metagraph model and its multicomponent and matrix representations. The proposed approach is based on a data model that combines incidence and nesting matrices, allowing metagraph operations to be formalized as matrix calculus. A formal description of the transformation algorithm is provided, complemented by a detailed application example and an interpretation of the resulting matrix elements. The proposed algorithm that integrates matrix-based metagraph representation with the event sourcing architectural pattern is expected to facilitate high-performance processing in data-intensive domains.

The annotating metagraph model is a powerful tool for describing complex systems with hierarchy and emergent properties, yet its syntactic representation poses a non-trivial challenge. The purpose of this work is to investigate the problem of representing this model using popular graph description languages: DOT, GraphML, and JSON Graph Format. A comparative analysis of their syntactic capabilities is conducted by encoding metagraphs that contain nested and overlapping metavertices. The research concludes that these languages face fundamental limitations for metagraph description. Their syntax, based on a strict tree-like structure, prevents the accurate representation of key metagraph features, such as an element’s membership in multiple, overlapping metavertices. To address this issue, the paper proposes a new, specialized format called MetagraphYAML. Its id-based referencing system fully supports all capabilities of the annotating metagraph model, thereby overcoming the limitations of existing approaches.

Graph visualization is a crucial tool for the visual analysis of interconnected data; however, traditional methods often fail to efficiently represent nested, multi-level hierarchical structures. This study proposes a method based on classical force-directed layout algorithms, specifically designed for visualizing metagraphs through a planetary interaction model of vertices. The distinctive feature of the proposed approach lies in incorporating the intrinsic weight of each vertex. This adaptation allows for the preservation of nested structures, improved graph readability, and scalability of the visualization as data complexity increases. Simulation results demonstrate the method's effectiveness across various types of nested data, including file systems, organizational hierarchies, and biological ontologies.

Existing semantic similarity scores are poorly suited to Russian language, because they compare sentences lexically rather than semantically. In the context of text generation problems in Russian, this problem is particularly relevant. This article is devoted to the development of a new evaluation that is as close as possible to human assessment and best fulfills the task of evaluating semantic similarity in the Russian language. The aim of the article is to develop a new automatic evaluation of the semantic similarity of two sentences for the task of generating texts in Russian. In the course of the work, a new evaluation has been developed based on existing sentence-transformers, trained first to understand the Russian language, and then to understand Russian semantics. The results suggest that the new inter-sentence semantic similarity score estimates the similarity between two sentences closest to a human assessment. The resulting assessment is of practical value for developers of question-and-answer and other systems that use text generation in Russian. The evaluation model is planned to be used in the task of generating texts in Russian.

Data preprocessing in machine learning tasks is an important step in the data mining process. To automate the data processing process and make it more suitable for the data under study, data preprocessing methods are implemented in AutoML systems. The purpose of the work is to compare the quality of work of AutoML systems for building a target model and training it. Study of the operation of modern AutoML systems has been conducted. Recommendations have been proposed for the use of unsupervised machine learning algorithms for the tasks of filling gaps, detecting and removing anomalies, and reducing the dimensionality of a data set. The conducted research allows us to determine the applicability of modern AutoML systems for building a machine learning model, to better understand the features of the systems, and to find out the possibility of their use in solving practical problems.