Hou Pong (Ken) Chan

Our source code separately save the word2idx and idx2word dictionary in the vocab.pt, they are not inside our saved model, so we still need to load it.

word2idx and idx2word are saved by the preprocessing scripts. I think you can still use the same command to get the predictions.

If you want to use catSeqD, the target sequences should have at least one since the orthogonal regularization loss requires the hidden states of the tokens.

You need to remove all the present keyphrases from the training data before running the preprocess.py. You need to remove the delimiter as well since you don't have present keyphrases.

I remembered that the loss is very small from the beginning.

Sorry I do not have any pre-trained models since it was more than three years. I remembered that the best checkpoint is usually located at the 3rd or 4th epoch....

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. * [Movie](https://www.dropbox.com/s/a7hp0ywtuyfkakz/movie_multi_view_output.tar.gz?dl=0) * [Sport](https://www.dropbox.com/s/7hq2lwlnn7mbs63/sport_multi_view_output.tar.gz?dl=0) *...

Yes, the `\n` is automatically generated by the model itself.

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...

Hi, is there any plan to release the code? Thanks