Scholarlink

This article presents the results of research on the effectiveness of various information retrieval methods in Retrieval-Augmented Generation (RAG) systems, with particular emphasis on their application in e-commerce. The research was conducted on a database containing information about products from an online shop, testing various search techniques based on embedding vectors, the BM25 method, a hybrid combining different approaches, and the use of a reranker. Attention was also paid to splitting documents into chunks and the use of the Conversational Query Reformulation (CQR) technique, which allows for the augmentation of the user's query. The experiments were conducted on various test questions regarding specific products and an overview of the shop's range of products. The effectiveness of each method was evaluated using precision, recall, and F1 metrics.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Hybrid Retrieval in RAG: A Comparison of Semantic, Lexical and Reranking Methods

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Marcin Gabryel,
  • Milan Kocić,
  • Aleksandra Gabryel

摘要

This article presents the results of research on the effectiveness of various information retrieval methods in Retrieval-Augmented Generation (RAG) systems, with particular emphasis on their application in e-commerce. The research was conducted on a database containing information about products from an online shop, testing various search techniques based on embedding vectors, the BM25 method, a hybrid combining different approaches, and the use of a reranker. Attention was also paid to splitting documents into chunks and the use of the Conversational Query Reformulation (CQR) technique, which allows for the augmentation of the user's query. The experiments were conducted on various test questions regarding specific products and an overview of the shop's range of products. The effectiveness of each method was evaluated using precision, recall, and F1 metrics.