Training data, model outputs and pretrained checkpoints for WWW 2021 paper ``Diverse and Specific Clarification Question Generation with Keywords" - data - contains the processed and cleaned *Home and Kitchen* dataset - data_office - contains the processed and cleaned *Office* dataset - output - contains all outputs on the test set of *Home and Kitchen* of the 7 compared systems in Table 4 - beam: KPCNet(-filter) - beam_filt: KPCNet(beam) - diverse_beam: KPCNet(divbeam) - hMup: hMup - kwd_cluster: KPCNet(cluster) - kwd_samples: KPCNet(sample) - MLE: MLE - the suffix like `beam`+X, where X indicates the order in beam search - hMup is an exception. X indicates the id of the *expert*. `beam4` achieved the best BLEU, and is thus used in both individual and group level evaluation. - the suffix for kwd_cluster and kwd_samples has an additional `a`+X, where X indicates the group id for conditioning keyword set - All variants of KPCNet has additional file with suffix `kwd_samples` - The words on the i-th line are conditioning keywords for the i-th sample - ckpt - pretrained model checkpoint - hparams - hyperparameters for the pretrained checkpoint