original source link: https://worksheets.codalab.org/worksheets/0xbda93e6519134c1ab1893ceaa19c8a5c/

processed_json: the files are downloaded from the above link, in the section right before the 'How well automatic metrics correlate with human judgement?'. 

raw_json: extracted from the data file, which can be found in the 'Code and data' Section in the above source link, towards the very beginning.

sorted_scores.json: built from processed_json. In this file, for each article, its summaries are sorted in the ascending order by their human rating scores.

doc_summ_bert_vectors.pkl: output of 'step1_encode_doc_summ.py'. This file will be used by 'step2_train_rewarder.py'.
