INTELLIGENT DECISION SUPPORT SYSTEMS FOR IDENTIFICATION OF CULTURAL VALUES: MODERN APPROACHES AND DEVELOPMENT PROSPECTS

Abstract

The relevance of the research is determined by the rapid digitalization of cultural heritage, the growing volume of digital images, textual descriptions, metadata, and the increasing need for accurate, adaptive, scalable, and explainable automated identification and classification of cultural heritage objects. The problem addressed in the study is the limited effectiveness of traditional and partially integrated approaches based on k-NN, manually designed descriptors, convolutional neural networks, and static knowledge graph representations when processing heterogeneous, incomplete, and semantically complex cultural heritage data. The purpose of the research is to analyze modern approaches to intelligent decision support systems for cultural heritage identification and to substantiate an improved multimodal architecture capable of integrating visual, textual, and semantic information. The research methods include comparative analysis of existing approaches, analysis and generalization of previous research results, conceptual modeling, and evaluation of modern deep learning, multimodal learning, and graph-based methods. Particular attention is given to Graph Neural Networks, Graph Attention Networks (GAT), Relational Graph Convolutional Networks (R-GCN), Vision Transformers, BERT-based language models, self-supervised learning, and Explainable AI. The results demonstrate that replacing static graph representation methods with GAT and R-GCN enables more effective consideration of relationship types and weights, contextual information, and incomplete or noisy data. A multimodal architecture integrating Vision Transformer for image processing, a BERT-based model for textual information, and GNNs for Knowledge Graph representation is proposed. Their integration into a shared embedding space enables cross-modal analysis, semantic search, and zero-shot and few-shot classification. The study concludes that the transition from isolated classification models to integrated multimodal intelligent decision support systems can significantly improve the accuracy, robustness, adaptability, and interpretability of cultural heritage identification. Further research should focus on adaptive decision-making, self-supervised learning, Explainable AI, automatic Knowledge Graph enrichment, integration with national and international cultural heritage databases, human-in-the-loop approaches, and federated learning.