mteb/TopiOCQA_validation_top_250_only_w_correct-v2

TopiOCQAHardNegatives An MTEB dataset Massive Text Embedding Benchmark TopiOCQA (Human-in-the-loop Attributable Generative Retrieval for Information-seeking Dataset) is information-seeking conversational dataset with challenging topic switching phenomena. It consists of conversation histories along with manually labelled relevant/gold passage. The hard negative version has been created by pooling the 250 top documents per query from BM25, e5-multilingual-large and e5-mistral-instruct.… See the full description on the dataset page: https://huggingface.co/datasets/mteb/TopiOCQA_validation_top_250_only_w_correct-v2.