Computer Tools in Education

Solving multicriteria problems of rating alternatives based on pairwise comparisons. Part II

2025-12-25T22:22:32+00:00

A number of well-known multicriteria problems of evaluating alternatives based on pairwise comparisons are considered. In these problems, given matrices containing results of paired comparisons of criteria and alternatives, one needs to find an absolute rating (priority, weight) of each alternative for decision making. Solutions to the problems are presented obtained using the method of analytical hierarchy process, the method of weighted geometric means, and the method of log-Chebyshev approximation of pairwise comparison matrices. The results obtained show that for some problems, solutions found by different methods may significantly differ from each other. In such cases, the decision to choose the best alternative may be based on additional analysis and comparison of the results of the problem solution obtained by all the methods used.

Comparison of Methods for Generating Synthetic Non-Stationary ECG-Like Signals for Testing Time Series Analysis Algorithms

2025-12-28T01:06:31+00:00

In this paper, various approaches to the generation of synthetic signals simulating a human electrocardiogram (ECG) are considered, with an emphasis on non-stationarity and the presence of various waveforms. Three main types of methods are proposed: 1) rule-based, based on the sum of Gaussians for modeling waves P, Q, R, S, T; 2) stochastic models based on Markov chains, allowing to emulate transitions between different physiological states; 3) neural network generators without strict rules (for example, a recurrent LSTM network with random weights). It is shown how each of the models can be modified to introduce nonstationarity (variations in the duration of cardiac cycles, switching states) and adding local recording artifacts (noisy areas). The proposed methods can be used in testing clustering and time series analysis algorithms when it is necessary to test the methods’ resistance to noise, rare events, and state changes.

Modelling a career guidance system using social network data analysis

2025-12-28T01:47:24+00:00

The paper addresses the problem of automating career guidance through the analysis of digital footprints of VKontakte users. The aim of the study is to enhance the accessibility and accuracy of diagnosing professional interests by means of the "AI Proforientator" mini-application on the VK Mini Apps platform. The methodology is based on a multi-label classification approach: each of the six RIASEC professional types is represented by a separate binary output with sigmoid activation, and the model is trained using the Binary Cross-Entropy loss function. Training data included users who had completed the Holland test, along with their community subscriptions collected through the Psychological Tests app. The technical implementation follows a client–server architecture with REST API, covering data preprocessing, interaction with the ML server, and result delivery. The developed system provides a full processing cycle: extraction and validation of subscriptions, probabilistic prediction of professional personality types according to Holland’s six categories, and presentation of results to the user. The model achieved a Top-1 accuracy of 47.1% and a Top-2 accuracy of 72.3%, confirming its applicability to career guidance diagnostics. The system automates the initial career guidance process and can be employed both by individual users and organisations to support educational trajectories, optimise recruitment, and improve staff development. The novelty of the work lies in the comprehensive integration of neural network methods for multi-label classification with a client–server application embedded in a social network, ensuring a personalised and scalable approach to career guidance.