Sentence-BERT (SBERT) is a version of BERT trained to turn whole sentences into embeddings, so similar sentences can be found by comparing vectors instead of rerunning the model on every pair.
In plain words
Plain BERT can judge whether two sentences mean the same thing, but only by reading both together. Finding the closest match in 10,000 sentences that way takes about 65 hours. SBERT trains BERT in a twin (siamese) setup so each sentence gets its own vector once. Comparing vectors with cosine similarity then takes seconds. The idea lives on in the Sentence Transformers Python library.
A quick example
A help desk has 10,000 past questions. With SBERT, each question is turned into a vector once. When a new question arrives, it is embedded and compared against all stored vectors, and the closest past questions appear almost instantly.
Where you'll meet it
- Semantic search and finding duplicate questions.
- Clustering and comparing sentences.
Related
Sources
- PaperSentence-BERT: Sentence Embeddings using Siamese BERT-Networks, Reimers and Gurevych (arXiv) · read 29 Sept 2026
- DocsSentence Transformers documentation, SBERT.net · read 29 Sept 2026
Checked by WhatIsAI against the sources above · 29 Sept 2026. Spotted an error? Report it.