Salient Phrase Aware Dense Retrieval: Can a Dense Retriever Imitate a Sparse One?
This addresses a key limitation in retrieval systems for applications requiring robust phrase matching and domain generalization, representing a novel hybrid approach rather than an incremental improvement.
The paper tackles the problem that dense retrievers lag behind sparse methods like BM25 in matching salient phrases and generalizing to out-of-domain data, by introducing SPAR, a dense retriever that imitates sparse models and achieves superior performance on tasks including five QA datasets, MS MARCO, and out-of-domain benchmarks.
Despite their recent popularity and well-known advantages, dense retrievers still lag behind sparse methods such as BM25 in their ability to reliably match salient phrases and rare entities in the query and to generalize to out-of-domain data. It has been argued that this is an inherent limitation of dense models. We rebut this claim by introducing the Salient Phrase Aware Retriever (SPAR), a dense retriever with the lexical matching capacity of a sparse model. We show that a dense Lexical Model Λ can be trained to imitate a sparse one, and SPAR is built by augmenting a standard dense retriever with Λ. Empirically, SPAR shows superior performance on a range of tasks including five question answering datasets, MS MARCO passage retrieval, as well as the EntityQuestions and BEIR benchmarks for out-of-domain evaluation, exceeding the performance of state-of-the-art dense and sparse retrievers. The code and models of SPAR are available at: https://github.com/facebookresearch/dpr-scale/tree/main/spar