Passage Re-ranking with BERT
Rodrigo Nogueira, Kyunghyun Cho
TL;DR
This paper demonstrates that re-implementing BERT for passage re-ranking significantly improves retrieval performance, achieving state-of-the-art results on major datasets.
Contribution
It presents a simple BERT-based re-ranking system that outperforms previous methods on TREC-CAR and MS MARCO datasets.
Findings
Achieved state-of-the-art results on TREC-CAR dataset.
Top entry in MS MARCO passage retrieval leaderboard.
Outperformed previous methods by 27% in MRR@10.
Abstract
Recently, neural models pretrained on a language modeling task, such as ELMo (Peters et al., 2017), OpenAI GPT (Radford et al., 2018), and BERT (Devlin et al., 2018), have achieved impressive results on various natural language processing tasks such as question-answering and natural language inference. In this paper, we describe a simple re-implementation of BERT for query-based passage re-ranking. Our system is the state of the art on the TREC-CAR dataset and the top entry in the leaderboard of the MS MARCO passage retrieval task, outperforming the previous state of the art by 27% (relative) in MRR@10. The code to reproduce our results is available at https://github.com/nyu-dl/dl4marco-bert
Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Code & Models
- 🤗amberoad/bert-multilingual-passage-reranking-msmarcomodel· 5.8k dl· ♡ 875.8k dl♡ 87
- 🤗xpmir/monobertmodel· 5 dl· ♡ 15 dl♡ 1
- 🤗sinequa/passage-ranker-v1-XS-multilingualmodel· 377 dl377 dl
- 🤗sinequa/passage-ranker-v1-XS-enmodel· 391 dl· ♡ 1391 dl♡ 1
- 🤗sinequa/passage-ranker-v1-L-multilingualmodel· 375 dl375 dl
- 🤗sinequa/passage-ranker-v1-L-enmodel· 376 dl376 dl
- 🤗sinequa/passage-ranker.chocolatemodel· 427 dl427 dl
- 🤗sinequa/passage-ranker.strawberrymodel· 322 dl322 dl
- 🤗sinequa/passage-ranker.mangomodel· 423 dl423 dl
- 🤗fluid-ai/bert-multilingual-passage-reranking-msmarcomodel· 8 dl· ♡ 98 dl♡ 9
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsTopic Modeling · Natural Language Processing Techniques · Multimodal Machine Learning Applications
MethodsLinear Layer · Cosine Annealing · Sigmoid Activation · Tanh Activation · Residual Connection · Attention Dropout · Linear Warmup With Linear Decay · Linear Warmup With Cosine Annealing · Byte Pair Encoding · Dense Connections