Closed michaelmoju closed 4 years ago
Sorry about the confusion. The term-based IR results are actually contained in the data/p_hotpotqa/hotpotqa_paragraph_level/*.jsonl
files. The qid field can be mapped to the original id in data/hotpotqa/*.json
files. If you can group each json line by qid, you can get the term-based IR results for each query in hotpotqa. The BERT model only give score for each term-based retrieved item. We will update the README later today regarding this issue.
Let me know if you have any other confusion or you can close this issue.
You have totally answered my question. Thank you!
Thanks for your contribution. I am wondering how can I get the intermediate term-based IR retrieval data? From the readme in your paragraph-level data, I assume the term-based results should be in
data/processed/content_selection_forward/hotpot_dev_p_level_unlabeled.jsonl
?