Introduction

This report provides a comprehensive evaluation of information retrieval (IR) systems, focusing on the performance of semantic search and its enhancements through combined methodologies. The study employs two core evaluation metrics—Recall, measuring completeness, and Normalized Discounted Cumulative Gain (NDCG), assessing the relevance and ranking of retrieved results. A novel assessment dataset was constructed using large language models (LLMs) to generate queries and rank document chunks, ensuring the dataset closely simulates real-world scenarios. The evaluation benchmarks semantic search against combined search strategies, further exploring reranking techniques to enhance retrieval precision and relevance. Key findings include improvements in recall and ranking through combined search methods and rerankers, highlighting the practical implications and opportunities for refining IR systems. The report concludes with insights on performance trade-offs and areas for further optimization.

This section presents research conducted by Unique on the following subjects