Unsupervised Corpus Aware Language Model Pre-training for Dense Passage Retrieval

TL;DR

This paper introduces coCondenser, an unsupervised pre-training method that enhances dense passage retrieval by improving robustness to data noise and reducing the need for large batches and extensive data engineering.

Contribution

It proposes coCondenser, a novel unsupervised pre-training approach that improves dense retriever training stability and performance without heavy engineering or large batch requirements.

Findings

coCondenser achieves comparable results to state-of-the-art systems

It reduces reliance on data augmentation and filtering

It enables effective training with small batches

Abstract

Recent research demonstrates the effectiveness of using fine-tuned language models~(LM) for dense retrieval. However, dense retrievers are hard to train, typically requiring heavily engineered fine-tuning pipelines to realize their full potential. In this paper, we identify and address two underlying problems of dense retrievers: i)~fragility to training data noise and ii)~requiring large batches to robustly learn the embedding space. We use the recently proposed Condenser pre-training architecture, which learns to condense information into the dense vector through LM pre-training. On top of it, we propose coCondenser, which adds an unsupervised corpus-level contrastive loss to warm up the passage embedding space. Retrieval experiments on MS-MARCO, Natural Question, and Trivia QA datasets show that coCondenser removes the need for heavy data engineering such as augmentation, synthesis, or filtering, as well as the need for large batch training. It shows comparable performance to RocketQA, a state-of-the-art, heavily engineered system, using simple small batch fine-tuning.