Generative Retrieval with Few-shot Indexing

TL;DR

This paper introduces Few-Shot GR, a training-free generative retrieval framework that uses prompt-based indexing with large language models to efficiently retrieve documents, outperforming traditional training-based methods.

Contribution

The paper proposes a novel few-shot, training-free indexing approach for generative retrieval using LLM prompts, reducing training costs and improving adaptability.

Findings

Few-Shot GR outperforms state-of-the-art training-based methods.

The approach effectively creates a docid bank without training.

Enhanced with one-to-many mapping, it further improves retrieval performance.

Abstract

Existing generative retrieval (GR) methods rely on training-based indexing, which fine-tunes a model to memorise associations between queries and the document identifiers (docids) of relevant documents. Training-based indexing suffers from high training costs, under-utilisation of pre-trained knowledge in large language models (LLMs), and limited adaptability to dynamic document corpora. To address the issues, we propose a few-shot indexing-based GR framework (Few-Shot GR). It has a few-shot indexing process without any training, where we prompt an LLM to generate docids for all documents in a corpus, ultimately creating a docid bank for the entire corpus. During retrieval, we feed a query to the same LLM and constrain it to generate a docid within the docid bank created during indexing, and then map the generated docid back to its corresponding document. Moreover, we devise few-shot indexing with one-to-many mapping to further enhance Few-Shot GR. Experiments show that Few-Shot GR achieves superior performance to state-of-the-art GR methods requiring heavy training.