A Cutting-Edge Intelligent Model for Robust Semantic Search Solutions

Semantic search and information Retrieval (IR) are fundamental components of modern artificial intelligence (AI) systems, offering effective automation of manual searching and relevant information finding. In this study, we propose IR technique based on semantic search results that links semantically similar text descriptions to retrieve information that often demand domain-specific knowledge. The first description typically represents a query, while the second is associated with additional detailed information to be linked. Our approach explores multiple techniques, including K-Nearest Neighbors (KNN), Retrieval-Augmented Generation (RAG), and Dual Encoder architecture. A comparative performance analysis was conducted of these methods where RAG peaked at 60% but struggled with similar inputs and the Dual Encoder model excelled in retrieval with Recall@5 of 93.18% and Top-1 accuracy of 73.56%. The best result was achieved by KNN, reaching 92.44% accuracy with Recall@5 of 100%, demonstrating its robustness in linking the semantic descriptions and its related information.