LED: Lexicon-Enlightened Dense Retriever for Large-Scale Retrieval

TL;DR

This paper introduces LED, a dense retrieval model enhanced with lexicon-aware representations, which aligns with a lexicon-aware teacher model through knowledge distillation, improving retrieval performance on benchmark datasets.

Contribution

The paper proposes a novel alignment method for dense retrievers using weakened knowledge distillation with lexicon-aware contrastive and rank regularization techniques.

Findings

LED outperforms baseline dense retrievers on three benchmarks.

The model surpasses its lexicon-aware teacher in retrieval effectiveness.

Combining LED with standard ranker distillation further improves results.

Abstract

Retrieval models based on dense representations in semantic space have become an indispensable branch for first-stage retrieval. These retrievers benefit from surging advances in representation learning towards compressive global sequence-level embeddings. However, they are prone to overlook local salient phrases and entity mentions in texts, which usually play pivot roles in first-stage retrieval. To mitigate this weakness, we propose to make a dense retriever align a well-performing lexicon-aware representation model. The alignment is achieved by weakened knowledge distillations to enlighten the retriever via two aspects -- 1) a lexicon-augmented contrastive objective to challenge the dense encoder and 2) a pair-wise rank-consistent regularization to make dense model's behavior incline to the other. We evaluate our model on three public benchmarks, which shows that with a comparable lexicon-aware retriever as the teacher, our proposed dense one can bring consistent and significant improvements, and even outdo its teacher. In addition, we found our improvement on the dense retriever is complementary to the standard ranker distillation, which can further lift state-of-the-art performance.