Database Indexing Using Deep Learning Algorithms

Summary. Automation of database indexing is a crucial component of modern database management systems that enhances search performance, scalability, and relevance in large-scale data environments. This paper explores the application of deep learning algorithms for building and optimizing vector indexes capable of automatic adaptation to changes in data structure and query patterns. An experimental comparison was conducted between traditional indexing methods (B-Tree, GIN in PostgreSQL) and vector-based indexing using Sentence-BERT embeddings implemented in FAISS and Milvus systems.

The study involved constructing a corpus of 1–5 million text records and performing a series of tests to evaluate performance metrics, including average query time, latency (95th percentile), index size, memory usage, and the search completeness metric (recall@10) relative to the BM25 baseline search. The results revealed that vector indexing achieved up to 0.94 recall@10 with comparable query latency, significantly outperforming traditional approaches in semantic search efficiency.

The practical contribution of this work lies in demonstrating the feasibility of hybrid DBMS architectures that combine traditional SQL indexing for fast filtering with vector-based methods for semantic refinement. This approach lays the foundation for the development of intelligent, self-optimizing databases capable of autonomous adaptation. Future research should focus on reinforcement learning for dynamic index restructuring, optimization of computational costs, and the integration of vector indexing into cloud-based and distributed systems.