This paper proposes a large language model (LLM)driven framework for classroom feedback scoring that integrates dual alignment mechanisms to ensure interpretability and fairness. The approach addresses long-standing concerns regarding the opacity of automated scoring by embedding semantic alignment through attention regularization and pedagogical alignment via rubric-based fine-tuning. Data were collected from over 65,000 classroom feedback entries spanning secondary and higher education contexts across three countries, yielding more than 7.3 million words of analyzed text. Extensive preprocessing safeguarded ethical compliance while preserving discourse structure. Experimental evaluation demonstrates significant improvements in prediction accuracy, robustness under rubric perturbations, and interpretability outcomes compared to baseline systems. Quantitatively, the framework reduces root mean square error by 21.3% relative to state-of-the-art Transformer models, while rubric coherence rises by 27.4%. Statistical analyses confirm improvements across all rubric dimensions, with effect sizes ranging from medium to large (Cohens d = 0.630.87). Teacher survey data further reveal that 82% reported higher trust in model outputs, while confirmatory factor analysis supports a three-dimensional construct of trust, pedagogical meaningfulness, and usability with high internal consistency (Cronbachs = 0.92). The findings demonstrate that accuracy and interpretability are not mutually exclusive but can reinforce one another, establishing a methodological foundation for transparent, scalable, and pedagogically aligned educational AI.

This study investigates the integration of learning analytics with fairness quantification in formative assessment, with an emphasis on contextualized interventions that respond to inequities in student learning processes. The research adopts a multi-institutional dataset comprising over 4,500 students across 18 classrooms, integrating log files, assessment records, and survey responses to ensure demographic, behavioral, and socio-cognitive diversity. A hybrid fairness quantification model is developed, combining statistical fairness metrics such as equal opportunity and disparate impact ratios with learning analytics indicators, including feedback latency, participation depth, and adaptive engagement. Interventions were designed through a three-layered protocol involving algorithmic detection of inequities, contextual mapping of student profiles, and targeted instructional adjustments. The results show that the fairness gap is most significant in feedback distribution, and the delay has a particularly severe impact on students from less affluent socioeconomic groups. The intervention measures increased the fairness index score by an average of 0.37 points on the standardized 0-1 scale, and the student satisfaction score was 21% higher than that of the control group. The benefits confirmed by the three-semester longitudinal follow-up were consolidated, and the standard deviation of the fairness index decreased from 0.18 to 0.07, indicating greater fairness consistency among the cohorts. Research has found that an analytical framework that emphasizes fairness not only enhances the transparency of formative assessment but also improves scalable, evidence-based intervention measures, thereby bringing about sustainable educational equity reforms.

O b j e c t i v e s. The article examines the features of using two-layer artificial neural network in problems of approximating binary functions of many binary variables. The issues of choosing the initial values of the model weights and choosing the number of neurons on the hidden layer are studied.M e t h o d s. The problem of approximating a binary function using an artificial neural network is reduced to the geometric problem of dividing the vertices of a multidimensional cube by hyperplanes. Combinatorial methods are used to prove lemmas on ways of dividing a hypercube by a hyperplane and to construct a lower estimate for the number of binary functions that can be approximated using one neuron on the hidden layer.R e s u l t s. The features of setting the initial values of weights of an artificial neural network are considered. A lower bound is constructed for the number of binary functions that can be approximated using an artificial neural network with one neuron on the hidden layer. The algorithmic complexity of calculating such an estimate is found. Numerical results are presented for using two-layer artificial neural networks to approximate binary functions in information security problems.C o n c l u s i o n. The results of the article allow choosing the parameters of an artificial neural network to improve the accuracy of approximation of binary functions of many variables.

O b j e c t i v e s. The aim of the work is to develop the architecture of an information system for transcription and translation of speech, implement its blocks and test their operation.M e t h o d s. The existing methods of speech recognition are considered; a comparative analysis of speech recognition and text translation models is carried out. The speech transcription process includes several successive stages: collection and preliminary processing of the audio signal, extraction of acoustic features, direct speech recognition, post-processing and text correction, and output of the result. At the stage of audio signal pre-processing, a combination of specialized libraries is used to prepare data for subsequent analysis. To normalize the recording parameters, the Librosa library is used, which allows resampling the signal to a standard frequency of 16 kHz and converting it to a monophonic format. To suppress background noise and highlight the speech component, the Demucs neural network model is used. The spectral subtraction algorithm additionally corrects residual noise. Speech activity segmentation (VAD) is performed using an energy detector from WebRTC, automatically highlighting speech fragments and removing pauses. The whisper-turbo (OpenAI) model was chosen to implement the speech recognition system due to the higher data processing speed, which allows implementing the streaming mode of the system, and lower requirements for the computing power of the machine. The translation module of the developed intelligent system is built on the T5-large-1024 (Text-to-Text Transfer Transformer) model, adapted for multilingual tasks.R e s u l t s. A method for creating an intelligent speech recognition system is proposed - a modular architecture of the speech recognition and translation system, a prototype is implemented and metrics are measured. The system showed the following results: for Russian-English translation Cosine Similarity 0.6951, WER 0.529, BLEU Score 0.239; for cascade Russian-Chinese translation through English Cosine Similarity 0.557, WER 0.748, BLEU Score 0.095. Research has shown that the use of cascade translation through English improves the quality of the final text by 32% according to the Cosine Similarity metric and by 25% according to BLEU Score compared to direct translation. The results of the implemented prototype were satisfactory.C o n c l u s i o n. The proposed implementation of the speech recognition system can solve the task with quality satisfactory for the described problem without risks of unauthorized access to data, since it works without an Internet connection. When using cascade translation through English, the quality of Russian-Chinese translation improves by 32% according to the Cosine Similarity metric (from 0.423 to 0.557) and by 25% according to BLEU Score (from 0.076 to 0.095). The proposed information system can be implemented in the educational process regardless of the academic discipline, and also used at exhibitions, conferences, and international forums. Parallel translation into different languages is possible, which will allow all participants of international forums to actively participate in its events.

O b j e c t i v e s. The objectives of the study are to collect data, develop an algorithm for automatic extraction of microexpressions from video recordings, and form rules for combinations of motor units, based on which basic human emotions are determined.M e t h o d s. Human facial microexpressions are brief, involuntary reactions that may appear when a person attempts to hide their true emotions. Microexpressions play a key role in lie detection and are an important indicator of the concealment of truthful information. In this article, Action Units (movement units) obtained using the py-feat library from the Facial Action Coding System (FACS) were used to analyse facial expressions.R e s u l t s. A dataset consisting of video recordings of a group of specific people was collected. Rules were developed based on combinations of action units and their intensities to determine basic emotions. An algorithm for determining and extracting microexpressions from video recordings was also formulated. The results of the algorithm study showed a negative correlation between the emotion of joy and the manifestation of lying.C o n c l u s i o n. The results obtained allow us to expand the information base for neural network lie detection using a video series of facial images by detecting and analysing microexpressions on them.

Here, we report the new strategy for the cationic RAFT polymerization of less reactive monomers such as p-methylstyrene and styrene. This strategy implies the addition of electron-withdrawing substituents into stabilizing group of cumyl dithiobenzoate that results in the decrease the stability of cationic intermediates thus facilitating their fragmentation and cationic RAFT polymerization. Particularly, it was demonstrated that both cumyl dithiobenzoate and cumyl dithiobenzoate with trifluoromethyl substituent in Z-group activated by small amount of SnCl4 induced living cationic RAFT polymerization of p-methoxystyrene affording polymers with molecular weight up to 80,000 and 30,000 g mol-1, respectively. In contrast to unsubstiuted counterpart, cumyl dithiobenzoate with trifluoromethyl substituent in Z-group was also efficient chain transfer agent for conducting of living cationic RAFT polymerization of p-methylstyrene giving polymers with Mn up to 10,000 g mol-1 and moderate dispersity (Đ= 1.3 – 1.8), while polymerization is terminated at incomplete conversion when cumyl dithiobenzoate was used. Cumyl dithiobenzoate with trifluoromethyl substituent in conjunction with SnCl4 also induced cationic RAFT polymerization of styrene affording well-defined polystyrenes with Mn up to 5,000 g mol-1. Finally, the block copolymers of p-methylstyrene with styrene and methyl methacrylate were sucesfully synthesized via mechanistic transformation from cationic to radical RAFT polymerization.

Submicron nanoparticles (NPs) of several semiconductors, such as Se or Te, exhibit pronounced dielectric Mie resonances in the visible range, which can be exploited for enhancing the efficiency of heterogeneous photocatalysis at the NP surface. Herein, it is demonstrated for the first time how dielectric Mie resonances in 190 nm amorphous a‐Se NPs and a‐Se/Ag2Se hetero‐NPs can enhance the efficiency of the model photocatalytic reactions involving organic (leuco‐methylene blue) and inorganic (K4[Fe(CN)6]) redox‐active species. Formation of the local enhanced electromagnetic field near a‐Se NPs surface governs efficient photogeneration of e‐h pairs either in a‐Se core or Ag2Se shell. The photoredox activity of a‐Se NPs depends on many factors and can be altered via formation of the secondary semiconductor shell. Particularly, formation of several nm thick Ag2Se shell on the surface of a‐Se NPs increases photocatalytic activity governed by dielectric Mie resonances in a‐Se/Ag2Se core‐shell NPs as compared to bare a‐Se NPs. It is believed that submicron a‐Se/MxSey (M = Ag, Cu, etc.) core‐shell NPs can be a promising platform for advanced semiconductor photocatalysts with enhanced activity in vis–NIR range and tunable redox properties.

The annual growth of meat production, accompanied by significant quality deterioration at all stages of the production chain, drives the development of fast and highly accurate control methods. The work is devoted to the adaptation of the spectrophotometric method for assessing pork quality based on the analysis of muscle tissue extracts. The purpose of the work is to generalize and systematize knowledge about spectrophotometric analysis and the application of this method for pork quality control during storage. The work provides a comparative spectrophotometric assessment of various methods for extracting protein and non-protein components of pork muscle tissue. Aqueous, buffer, NaCl and KCl extracts of muscle tissue were studied, their absorption spectra in the wavelength range of 315–1000 nm were analyzed. It was found that KCl and NaCl extraction ensured the maximum degree of myofibrillar and sarcoplasmic protein extraction, and also formed the most pronounced and stable spectral peaks. Particular attention was paid to the analysis of KCl extracts demonstrating the best resolution and clarity of spectral curves, which is important for a detailed study of changes in muscle tissue properties during storage. During meat storage, statistically significant changes in the intensity and geometry of key spectral peaks (λ325–335, λ 355, λ410–415, λ545, λ580, λ610–620, λ635–650) were revealed, which were simultaneous with histostructural transformations of muscle tissue. A high correlation was established between the change in the area of minor peaks and the dynamics of muscle fiber diameter, which allows using spectral characteristics as objective indicators for the degree of changes in muscle tissue at the cellular and molecular levels during storage. The results obtained confirm the feasibility of using spectrophotometric analysis of KCl extracts for an objective assessment of meat quality and monitoring its changes at various stages of storage.