dual_view_review_sum
dual_view_review_sum copied to clipboard
kenchan0226
Generated Summary
Hello, thanks for your great work ! I am wondering would you please shared the generated summaries for rouge calculation or model checkpoint ? Thanks so much !
The followings are our generated summaries for each dataset. Each zip file contains five sets of summaries generated by our model using different random seeds.
Hi, thanks so much for fast response!
After looking through the home_min_4_multi_view_multi_task_basic_DecStateIn_2enc_residual_2multi_hop_g0.8_c0.1_i0.1_gdp0.0_check1k_joint_stop_h512_seed_520.20191004-111942, I have a question about the generated summary: there are 110 summaries that contain more than one line. Is this \n symbol automatically generated by the model itself ?
so you make \n in your vocabulary and create ground truth by '\n'.join(summary_ls) when training?( since the summary format in the dataset is a list of string.)
A reference summary in the datasets is simply stored as a string and some of the reference summaries contain the \n token. We did not manually add any \n token into the reference summary.
The \n token is in our vocabulary since we extract the most frequent 50k words in the training set as our vocabulary.
During inference, we split the predicted summary (and the ground-truth summary) into a list of sentences because it is required by the pyrouge script.