Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks

Abstract

BERT (Devlin et al., 2018) and RoBERTa (Liu et al., 2019) has set a new state-of-the-art performance on sentence-pair regression tasks like semantic textual similarity (STS). However, it requires that both sentences are fed into the network, which causes a massive computational overhead: Finding the most similar pair in a collection of 10,000 sentences requires about 50 million inference computations (~65 hours) with BERT. The construction of BERT makes it unsuitable for semantic similarity search as well as for unsupervised tasks like clustering. In this publication, we present Sentence-BERT (SBERT), a modification of the pretrained BERT network that use siamese and triplet network structures to derive semantically meaningful sentence embeddings that can be compared using cosine-similarity. This reduces the effort for finding the most similar pair from 65 hours with BERT / RoBERTa to about 5 seconds with SBERT, while maintaining the accuracy from BERT. We evaluate SBERT and SRoBERTa on common STS tasks and transfer learning tasks, where it outperforms other state-of-the-art sentence embeddings methods.

Full paper on arXiv

Details

arXiv identifier

1908.10084

Published

2019-08-27

Authors

Nils Reimers, Iryna Gurevych

Models That Cite This Paper

• Described byparaphrase-multilingual-MiniLM-L12-v2
• Described byparaphrase-multilingual-mpnet-base-v2
• Described bybge-base-en-v1.5-course-recommender-v5
• Described byparaphrase-mpnet-base-v2
• Described byparaphrase-MiniLM-L6-v2
• Described bydistiluse-base-multilingual-cased-v2
• Described bydistiluse-base-multilingual-cased-v1
• Described byparaphrase-MiniLM-L3-v2
• Described bynli-mpnet-base-v2
• Described byparaphrase-albert-small-v2