samchain/BIS_speeches_97_23_MLM

Dataset Card for "BIS_Speeches_97_23" This dataset is built from scrapped speeches on the Bank of International Settlements thanks to this repo : https://github.com/HanssonMagnus/scrape_bis. The dataset is made of 12k speeches from 1997 to 2023. Each pair is built with extracted sentences from speeches, if B is following A then the 'next_sentence_label' is 1 else it is 0. Negative pairs are built by choosing a sentence from another speech randomly. Credits Full… See the full description on the dataset page: https://huggingface.co/datasets/samchain/BIS_speeches_97_23_MLM.