Domain-matched Pre-training Tasks for Dense Retrieval
This addresses the bottleneck of pre-training effectiveness for information retrieval, which is a domain-specific problem for NLP researchers and practitioners.
The paper tackles the problem of poor performance from additional pre-training in information retrieval by showing that domain-matched pre-training tasks can overcome this barrier. They pre-trained bi-encoder models on 65 million synthetic questions and 200 million Reddit post-comment pairs, achieving substantial improvements over supervised baselines on retrieval benchmarks.
Pre-training on larger datasets with ever increasing model size is now a proven recipe for increased performance across almost all NLP tasks. A notable exception is information retrieval, where additional pre-training has so far failed to produce convincing results. We show that, with the right pre-training setup, this barrier can be overcome. We demonstrate this by pre-training large bi-encoder models on 1) a recently released set of 65 million synthetically generated questions, and 2) 200 million post-comment pairs from a preexisting dataset of Reddit conversations made available by pushshift.io. We evaluate on a set of information retrieval and dialogue retrieval benchmarks, showing substantial improvements over supervised baselines.