LED: Lexicon-Enlightened Dense Retriever for Large-Scale Retrieval
This work addresses a specific bottleneck in large-scale retrieval for applications relying on semantic search, offering incremental improvements over existing dense retrieval methods.
The paper tackled the problem of dense retrievers overlooking local salient phrases and entity mentions in first-stage retrieval by aligning a dense retriever with a lexicon-aware representation model using weakened knowledge distillations, resulting in consistent and significant improvements on three public benchmarks, even outperforming the teacher model.
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.