neemakot / health-fact-checking Goto Github PK

Hi, thanks for creating this dataset!

In sec. 4.1 of your paper, it looks like you use a pipeline where you select relevant sentences from an evidence document, and then use BERT to predict the relation between the selected sentences and the claim. Does the main_text field in the data you make available for download correspond to the input evidence document?

What exactly is the relationship between the main_text and the sources? Is the main_text just the concatenation of the text from all the sources - and if so, what's going on in the cases where there is no source listed?

Thanks for the clarification!

Dave

Dear author,
Thanks for your great work. However, I find bug when I try to load dev.tsv. I use pd.read_csv and the length of the data is 1221 rather than 1227 in the README. Could you please provide some code about loading the dataset?