Understanding BERT Rankers Under Distillation
This addresses deployment challenges in real-world search scenarios for users needing efficient retrieval systems, though it is incremental as it builds on existing distillation methods.
The paper tackled the high computational cost of BERT-based rankers in information retrieval by transferring knowledge to a smaller model through distillation, achieving up to nine times speedup while preserving state-of-the-art performance.
Deep language models such as BERT pre-trained on large corpus have given a huge performance boost to the state-of-the-art information retrieval ranking systems. Knowledge embedded in such models allows them to pick up complex matching signals between passages and queries. However, the high computation cost during inference limits their deployment in real-world search scenarios. In this paper, we study if and how the knowledge for search within BERT can be transferred to a smaller ranker through distillation. Our experiments demonstrate that it is crucial to use a proper distillation procedure, which produces up to nine times speedup while preserving the state-of-the-art performance.