Patent SBERT-adapt-ub outperforms in patent similarity, highlighting the importance of training phase in embedding models.
