articles
| This article proposes a model for automatically generating educational tests based on domain-adaptive retraining of large language models (LLM). Traditional methods for developing test items are time-consuming and limited to narrow subject specialization. An analysis of existing approaches to generating test items using solely prompting of pre-trained LLMs revealed key limitations: unstable results, insufficient control over question structure, and the need for post-editing, which justifies the need to develop specialized solutions. A pipeline for retraining the T-Lite 1.0 language model (7 billion parameters) using the LoRA technique on a dataset of 4000 validated test tasks in higher education disciplines is proposed. A distinctive feature of the proposed method is the use of chain-of-thought to structure the task generation process by decomposing it into components: topic, goal, format, prerequisite knowledge, wording and expected response. An inference pipeline for the task generation system was developed, integrating multimodal processing of input data (text/image), automatic content segmentation, and multi-stage task generation using specialized Qwen2-VL and Gemma-2-27b-it models. The model was tested and implemented into the independent assessment ecosystem of the i-exam.ru portal. This expanded the functionality of existing online testing services by providing additional capacity to expand the uncompromised database of test-based assignments, which is particularly important for ensuring the reliability of online testing procedures, such as those of the FEPO and FIEB. This implementation confirms the model’s ability to generate high-quality educational tests that meet psychometric requirements and are suitable for use in the educational process. Continue... | |
|
№ 3(123)
30 june 2026 year
Rubric: Performance management Authors: Kalayda S., Masyutin S., Vorotilova M., Zubareva V. |
In the context of intensive growth in e-commerce volumes and an increase in the number of returns, the task of improving the efficiency of logistics management, including both direct and reverse flows, is becoming relevant. Reversible processes are particularly complex, due to the need to determine the real return reasons, which directly determine the choice of route nodes. The aim of the article is to develop a multi-model method for situational managing logistics flows in e-commerce, characterized by a comprehensive solution to the interrelated tasks of analysis of customer requests for returns and multi-criteria routing of material flows. To classify the return reasons, an intuitionistic fuzzy random forest was used, since it allows for the uncertainty, incompleteness, and inconsistency of customer data to be taken into account by using intuitionistic fuzzy sets and an ensemble of decision trees. To optimize routes in both forward and reverse logistics, the earthworm algorithm was used, thanks to a balanced approach to global and local search using two mechanisms of reproduction and Cauchy mutation. The proposed multi-model method is implemented as a software in the Python and integrated with corporate information systems and data warehouse. The results of his testing showed that the combined use of a text data processing tools, intuitionistic fuzzy random forest and an earthworm algorithm allows us to the development of an effective logistics management system. Customer request analysis and route optimization reduce operating costs for processing returns, while accumulated statistics on their causes provide the basis for proactively adjusting product policies and increasing the redemption rate. Continue... |
| The effectiveness of modern cities’ functioning depends on the level of development and connectivity of large spatially distributed systems, such as built-up spaces, road networks, engineering infrastructures, and information-communication networks. The consistency of structures of these systems directly affects residents’ access to infrastructure facilities. Urban SDS structures evolve under the influence of natural, historical, and anthropogenic factors, demonstrating signs of statistical self-similarity characteristic of stochastic fractal objects. Therefore, applying methods from fractal geometry to study the dynamics of development and self-organization of urban SDS has significant potential for planning and modeling transformations of city territories. Classical approaches of fractal analysis are focused on studying general patterns and global characteristics of SDS at the city-wide scale. Such data is useful for strategic and tactical planning but insufficient for forming an objective picture at the intra-city territorial units level. To overcome these limitations, it is proposed to perform spatial discretization of the studied SDS and obtain two-dimensional maps of spatial fractal data that can be effectively mapped. Building upon the cartograms implemented in the DCFA method, a technology for creating schematic maps is proposed, which reflect the distribution of quantitative assessments of fractal dimensionality in the studied urban SDS. For quantitatively assessing consistency, a difference map between spatial datasets of the investigated systems is suggested. Additionally, a procedure for calculating an effective sequence of cell sizes for implementing the box-counting algorithm is presented. An example of analyzing the coherence between road networks and urban development in Zelenograd using the proposed procedure is provided. Continue... | |
| A game-theoretic model of oligopoly with reflexive player behavior and quantity conjectural variations is proposed. The model generalizes two types of player reflexion –symmetric and asymmetric. It allows to describe the forecasts regarding competitors’ actions more flexibly and to study a wider range of market equilibrium options. For symmetric reflexions, the model is examined in terms of the influence of reflexivity rank and the number of players, revealing the dependence of equilibria on the depth of players’ strategic thinking. For asymmetric reflexions, the model is analyzed depending on the type of representation and the number of players. The goal of the study is to develop a method for calculating conjectural variations for both symmetric and asymmetric player representations. Unlike existing methods for calculating conjectural variations, which ignore the interactions between the number of players and the depth and type of reflexivity, this study yields new results for linear demand and cost functions. These results have significant practical implications and allow for the calculation of game equilibria with arbitrary reflexivity ranks and an arbitrary number of players, while also taking into account the type of reflexion of the analyzed player. This expands the toolkit for modeling real-world market situations. Thus, this study makes a significant contribution to the development of reflexive game theory by offering an adequate mathematical framework for analyzing strategic interactions in oligopoly settings, taking into account the heterogeneous expectations of players, which is necessary for forecasting market dynamics. To visualize the analytical results, graphical interpretations of the patterns of change in hypothetical variations are presented, which are used to formulate conclusions. Continue... | |
| When considering the motion of statically unstable objects, such as a cyberphysical system in the form of a seven-link exoskeleton mechanism, the problem of adhesion to the supporting surface plays a significant role during anthropoid walking. This is because operating conditions change during movement: gait phases alternate, the characteristics of the supporting surface change, and external disturbances occur. To improve stability and prevent slippage, it is necessary to consider the characteristics of foot motion. Therefore, the presence of weighty inertial feet ensures the novelty and relevance of the proposed model. This paper presents a model of a cyber-physical system for an exoskeleton mechanism moving in a vertical plane and creating the prerequisites for implementing a situational approach to movement control. The model includes seven moving links connected by hinges, ensuring the maximum approximation of the model to a real exoskeleton. The objective of the study is to develop and describe algorithms for designing motion control for an exoskeleton mechanism using a software method, taking into account situational parameters, applying a previously developed method for controlling a robotic anthropoid mechanism in the form of programmable movements. The testing of the developed algorithms was carried out in the Wolfram Mathematica environment using mathematical modeling tools, numerical methods and visualization. In the course of the work, a model of an exoskeletal mechanism with seven movable links was developed, including weighty feet moving in a vertical plane. The originality of the development lies in the use of angles between the links when describing the model and the use of rotation matrices of the coordinate axes when composing a system of differential equations. A mathematical model of the presented link of the exoskeletal mechanism and algorithms for solving direct and inverse dynamics problems for it have been developed. A motion visualization routine has been created that allows you to demonstrate the motion process of the model in question in accordance with the specified functions. The necessity of taking into account situational parameters when using an exoskeleton is shown. The relevance of using fuzzy logic methods and neural networks to optimize modeling while adapting to environmental uncertainty is described. Continue... | |
| In the Interval Method of Target Solution Displacement the problem formulation can be refined by an expert at any step of the interactive solution search process, based on the analysis of intermediate results and situation portraits (target, current, and achieved). The best approximation to a feasible solution can be found even with an inconsistent system of constraints. At all stages of computation, IMTSD operates not with boundaries but with real-number segments as integral objects. IMTSD allows the integration of domain expert knowledge (represented in the form of mandatory and guiding requirements for the solution) with a formalized step-by-step search for feasible solutions. The step-by-step human-in-the-loop solution search scheme enables the expert (or an intelligent robot) to: analyse constraint violations; evaluate the feasibility and effectiveness of solutions; input directives that guide the search (“increase”, “decrease”, “fix” the right-hand sides of selected constraints); adjust the priorities of selected constraints and the levels of applied data and solution precision; modify the problem formulation by adding/removing constraints, changing the number of main variables, and updating constraint coefficients. Numerical experiments have demonstrated the high efficiency of IMTSD in solving complex situational planning problems under uncertainty. An example of using IMTSD to solve the resource allocation problem in emergency situation is presented. IMTSD is considered as a contribution to the methodological arsenal of technologies for solving practically significant linear situational planning problems. Continue... |
