Published March 7, 2021 | Version v1
Software Open

Resource for WWW 2021 paper "Diverse and Specific Clarification Question Generation with Keywords"

  • 1. Shanghai Jiao Tong University

Description

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

Files

Files (234.6 MB)

Name Size Download all
md5:6d72be7053b9818c6af49a46b7f5dd79
67.1 MB Download
md5:a99c572048fdd69161ae5d9ac60f09f1
67.1 MB Download
md5:f609a564229ae48eae8e71e478d4a529
67.1 MB Download
md5:cc2ba9a6b493467166f69ac60ea4fa3b
33.3 MB Download

Additional details

Related works

Is supplement to
Conference paper: 10.1145/3442381.3449876 (DOI)