Search

Search Results

• A NEURAL NETWORK-BASED METHOD FOR EXTRACTING CAUSAL RELATIONS USING COMPARATIVE ANALYSIS OF SYNTHETIC AND OPEN CORPORA

  А.N. Tselykh , I. А. Valukhov

  160-169

  2026-09-10

  Abstract ▼

  The automatic extraction of causal relations is critically important for decision support systems, but its development is hindered by the scarcity of annotated corpora. The aim of this study is to comparatively analyze the effectiveness of three types of training data: open expert corpora, a manually annotated political corpus, and synthetic data generated by a large language model. Experiments were conducted using the DistilBERT architecture with token-level BIOES tagging. Synthetic data were generated with the Llama-3 LLM by encapsulating causes and effects in XML markers, followed by deterministic conversion into the BIOES format. For comparison, two expert corpora (EventStoryLine and SemEval-2010 Task 8), the expert-annotated political corpus PolitiCAUSE, and a synthetic sample of 1,100 balanced examples were used. The model trained on synthetic data achieved a Macro F1 score of 0.736, which is 6.7% higher than the result obtained from training on the expert corpora (0.690). Training on the political corpus PolitiCAUSE (24,417 examples) yielded a substantially lower result, with a Macro F1 score of 0.367. This discrepancy is attributed to class imbalance (approximately 15% causal tokens versus 50% in the synthetic sample) and to differences in task formulation: PolitiCAUSE is oriented toward binary sentence-level classification of causal presence, whereas the present study addresses token-level span labeling. The hypothesis that annotation quality and class balance are more important than data volume is confirmed. The proposed LLM-generation pipeline makes it possible to create training datasets that outperform expert corpora without direct manual annotation costs. The method is recommended for rapid deployment of causal relation extraction systems in new domains.

• LOW-RESOURCE ABSTRACTIVE TEXT SUMMARIZATION VIA CONTRASTIVE UNSUPERVISED LEARNING WITH A ROUGE-ORIENTED LOSS FUNCTION

  I. Е. Lysenko

  91-101

  2026-09-10

  Abstract ▼

  The relevance of this work is motivated by the fact that in specialized domains and low-resource languages obtaining a sufficient number of “document-summary” pairs for effective abstractive text summarization is expensive and often practically infeasible, whereas modern deep learning models require large labeled corpora and large volumes of text remain unused. The aim of this study is to develop a new training method for abstractive summarization models in low-resource (10-shot and 100-shot) settings that improves quality by fine-tuning the model on unlabeled data using unsupervised contrastive learning with input augmentation. The research tasks include designing a new contrastive loss function and comparing the proposed approach with existing methods of low-resource abstractive summarization. The methods and approaches comprise a new loss function for fine-tuning a transformer that includes a contrastive generative component based on a differentiable approximation of the ROUGE-3 metric. Two variants of the method are proposed – the sequential “DiffROUGE-seq” and the semi-supervised “DiffROUGE-sim”. BART-large is used as the base model, while input augmentations are generated by FLAN-T5-large in a zero-shot regime. Experiments are conducted on popular datasets AESLC, Gigaword, XSum, and Reddit. The proposed method achieves substantial improvements in ROUGE scores in successful cases, with average gains of 2.69 in the 10-shot setting and 2.13 in the 100-shot setting. In summary, the newly proposed low-resource abstractive summarization method that leverages both labeled and unlabeled data is significantly more effective in terms of ROUGE than existing approaches

• ANALYSIS OF ARTIFICIAL INTELLIGENCE METHODS APPLIED TO SOLVING PSYCHIATRY PROBLEMS

  E.S. Podoplelova

  2022-05-26

  Abstract ▼

  The use of artificial intelligence methods in the field of medicine has become widespread,
  helping to diagnose, analyze and make recommendations for treatment. Psychiatry is a branch of
  medicine that studies mental disorders, methods for their diagnosis and treatment. Her range of
  tasks includes not only diagnosis and treatment, but also observation, monitoring and subsequent
  rehabilitation of patients. This subject area has significant problems, such as objectivity, inconsistency
  in the diagnosis, the complexity of the classification of diseases, and the unpredictability
  of the course of the disease. With a number of these problems, the use of machine learning methods
  and artificial intelligence algorithms helps to cope. This paper is devoted to a review of research
  on artificial intelligence methods used to solve problems in the field of psychiatry.
  The relevance of the topic is due to the high need for improvements in this subject area. Specific
  issues are presented in this article. Among them, the main directions were identified: data deidentification,
  classification of symptom severity, accuracy of condition prediction. To solve them,
  the authors used such methods as latent semantic analysis for natural language processing, classification
  methods, convolutional neural networks for prediction, and cognitive modeling. Separately,
  the effectiveness of hybrid systems, including the implementation of several machine learning
  methods at once, is noted. The aim of the study was to highlight the main directions of development
  of research in the scientific community, which demonstrate the successful integration of artificial intelligence into psychiatry, as well as to compare them with each other according to the
  obtained estimates of the accuracy of the models. Which, in turn, implies the analysis and analysis
  of specific algorithms, their performance for specific tasks

• DEEP LEARNING METHODS FOR NATURAL LANGUAGE TEXT PROCESSING

  V.V. Kureichik, S.I. Rodzin, V.V. Bova

  2022-05-26

  Abstract ▼

  The analysis of approaches based on deep learning (DL) to natural language processing
  (NLP) tasks is presented. The study covers various NLP tasks implemented using artificial neural
  networks (ANNs), convolutional neural networks (CNNs) and recurrent neural networks (RNNs).
  These architectures allow solving a wide range of natural language processing tasks that previously
  could not be effectively solved: sentence modeling, semantic role labeling, named entity
  recognition, answers to questions, text categorization, machine translation. Along with the advantages
  of using CNN to solve NLP problems, there are problems associated with a large number
  of variable network parameters and the choice of its architecture. We propose an evolutionary
  algorithm for optimizing the architecture of convolutional neural networks. The algorithm initializes
  a random population of a small number of agents (no more than 5) and uses the fitness function
  to get estimates of each agent in the population. Then a tournament selection is carried out
  between all agents and a crossover operator is applied between the selected agents. The algorithm
  has such an advantage as the small size of the network population, it uses several types of CNN
  layers: convolutional layer, maximum pooling layer (subdiscretization), medium pooling layer and
  fully connected layer. The algorithm was tested on a local computer with an ASUS Cerberus Ge-
  Force ® GTX 1050 Ti OC Edition 4 GB GDDR5, 8 GB of RAM and an Intel(R) Core(TM) i5-4670
  processor. The experimental results showed that the proposed neuroevolutionary approach is able
  to quickly find an optimized CNN architecture for a given data set with an acceptable accuracy
  value. It took about 1 hour to complete the algorithm execution. The popular TensorFlow framework
  was used to create and train CNN. To evaluate the algorithm, public datasets were used:
  MNIST and MNIST-RB. The kits contained black-and-white images of handwritten letters and
  numbers with 50,000 training samples and 10,000 test samples.

• IDENTIFICATION OF KEY TECHNOLOGIES BASED ON COLLECTION AND ANALYSIS OF DATA FROM OPEN RUSSIAN-LANGUAGE SOURCES

  А.G. Bondarenko , А.G. Kravets

  144-159

  2025-07-24

  Abstract ▼

  This article is devoted to the development and approbation of a new approach to the collection, processing and analysis of open data in the Russian language for identification of key technological trends. To solve the problem of formation and subsequent analysis of structured datasets methods of web scraping, natural language processing and analysis of time-series have been developed and implemented via programming. The approach described in the article has been applied for the first time in order to extract and structure information from scientific articles, news resources and patent documentation in the Russian language for the first time. As a result of analyzing the obtained dataset of scientific publications, 30 most frequently mentioned bigrams and the same number of trigrams of technological terms have been identified. Based on the frequency analysis of bigrams and trigrams, key technological terms were   identified which then were used for complex filtration on key technologies. Complex filtration enabled to fulfill the search of patents in Russian and their collection for further analysis. As a result of preprocessing of the obtained patent data time series of patent activity have been formed. The programme system of key technological identification has been implemented in JavaScript and Python using Selenium and BeautifulSoup libraries for web scraping, NLTK and Scikit-learn for text data processing and analysis. The study focused on the dynamics of the development of key technologies over time has allowed to identify periods of intensive patent activity and declining interest in this or that kind of technology. The results presented in the article provide a basis for further development of machine learning methods for the purpose of predicting technological development and identifying promising areas of applied research.