InPars: Data Augmentation for Information Retrieval using Large Language Models

TL;DR

This paper introduces InPars, a method that uses large language models to generate synthetic data for information retrieval, improving zero-shot transfer and outperforming existing baselines.

Contribution

It presents a novel approach leveraging large language models for synthetic data generation to enhance IR task performance and transferability.

Findings

Models trained on synthetic data outperform BM25 and recent dense retrieval methods.

Retrievers trained on combined supervised and synthetic data achieve superior zero-shot transfer.

Synthetic data improves IR performance across various domains.

Abstract

The information retrieval community has recently witnessed a revolution due to large pretrained transformer models. Another key ingredient for this revolution was the MS MARCO dataset, whose scale and diversity has enabled zero-shot transfer learning to various tasks. However, not all IR tasks and domains can benefit from one single dataset equally. Extensive research in various NLP tasks has shown that using domain-specific training data, as opposed to a general-purpose one, improves the performance of neural models. In this work, we harness the few-shot capabilities of large pretrained language models as synthetic data generators for IR tasks. We show that models finetuned solely on our unsupervised dataset outperform strong baselines such as BM25 as well as recently proposed self-supervised dense retrieval methods. Furthermore, retrievers finetuned on both supervised and our synthetic data achieve better zero-shot transfer than models finetuned only on supervised data. Code, models, and data are available at https://github.com/zetaalphavector/inpars .