A Contrastive Pre-training Approach to Learn Discriminative Autoencoder for Dense Retrieval

TL;DR

This paper introduces a contrastive pre-training method for autoencoders with a lightweight decoder to enhance discriminative text representations for dense retrieval, outperforming existing models.

Contribution

It proposes a novel contrastive pre-training approach with a non-autoregressive decoder to improve the discriminative ability of autoencoder-based language models for dense retrieval.

Findings

Significantly outperforms state-of-the-art autoencoder models in dense retrieval tasks.

The contrastive strategy suppresses common words and emphasizes key words in decoding.

The method improves the quality of text representations for information retrieval.

Abstract

Dense retrieval (DR) has shown promising results in information retrieval. In essence, DR requires high-quality text representations to support effective search in the representation space. Recent studies have shown that pre-trained autoencoder-based language models with a weak decoder can provide high-quality text representations, boosting the effectiveness and few-shot ability of DR models. However, even a weak autoregressive decoder has the bypass effect on the encoder. More importantly, the discriminative ability of learned representations may be limited since each token is treated equally important in decoding the input texts. To address the above problems, in this paper, we propose a contrastive pre-training approach to learn a discriminative autoencoder with a lightweight multi-layer perception (MLP) decoder. The basic idea is to generate word distributions of input text in a non-autoregressive fashion and pull the word distributions of two masked versions of one text close while pushing away from others. We theoretically show that our contrastive strategy can suppress the common words and highlight the representative words in decoding, leading to discriminative representations. Empirical results show that our method can significantly outperform the state-of-the-art autoencoder-based language models and other pre-trained models for dense retrieval.