ucsahin/TR-Extractive-QA-82K

The dataset consists of nearly 82K {Context, Question, Answer} triplets in Turkish. Since most of the answers are only a few words and taken directly from the provided context, it can be better used in in finetuning encoder-only models like BERT for extractive question answering or embedding models for retrieval. The dataset is a filtered and combined version of multiple Turkish QA-based datasets. Please use ucsahin/TR-Extractive-QA-5K for more detailed and sampled version of this dataset.