DESIGN AND IMPLEMENTATION OF AN INTELLIGENT SEARCH SYSTEM BASED ON NEURAL NETWORKS

Background. In the digital information era, the ability to retrieve relevant data quickly and accurately is increasingly critical. Traditional search engines such as Google or Bing rely on keyword matching, which can fail in cases of vague queries, multilingual content, or media-based searches. The rapid development of neural networks and AI technologies introduces new opportunities to enhance search systems by understanding context, semantics, and user behaviour. This study aims to develop a search system based on ElasticSearch, integrating multiple neural network modules to improve search precision, personalisation, and flexibility. Methods. The proposed system includes four main components: ElasticSearch for full-text indexing, a convolutional neural network for image recognition, a graph-based semantic model for query expansion, and a ranking model based on historical user interactions. The backend is developed in Python using Visual Studio, with modular AI components that can be activated or disabled by the user. The semantic model represents terms as graph nodes and semantic proximity as weighted edges, enabling dynamic context-driven query refinement. Additional features include synonym detection, citation filtering, and user-specific ranking. Results and Discussion. Two key experiments were performed. The first examined system performance by testing search speed across database sizes ranging from 100 to 100,000 records. It was found that even with all neural modules enabled, latency remained minimal, confirming system scalability. The second experiment assessed the impact of training data on the quality of the semantic model. A model trained on low-quality, AI-generated data resulted in incoherent word associations and poor query expansion. In contrast, a model built on human-curated texts produced clear, logical semantic links and significantly improved search relevance. The image search function demonstrated the system’s ability to identify relevant visual content based on vague or partial user input, while the context expansion model enhanced result diversity and accuracy even with incomplete or ambiguous queries. Conclusion. This work presents a hybrid search engine that effectively integrates traditional indexing with AI-powered features. The system offers robust text and image search capabilities, intelligent semantic understanding, and personalised ranking. Experiments confirmed its efficiency, relevance, and adaptability across varying data conditions and resource levels. With modular architecture and advanced context handling, the system addresses limitations of conventional search engines and sets a strong foundation for future development in intelligent information retrieval. Keywords : Search system, search accuracy, neural networks, ElasticSearch.