summscreen

• Description:

SummScreen Summarization dataset, non-anonymized, non-tokenized version.

Train/val/test splits and filtering are based on the final tokenized dataset, but transcripts and recaps provided are based on the untokenized text.

There are two features:

• transcript: Full episode transcripts, each line of dialogue separated by newlines

• recap: Recaps or summaries of episodes

• Homepage: https://github.com/mingdachen/SummScreen

• Source code: tfds.datasets.summscreen.Builder

• Versions:

  • 1.0.0 (default): Initial release.

• Download size: 841.27 MiB

• Supervised keys (See as_supervised doc): ('transcript', 'recap')

• Figure (tfds.show_examples): Not supported.

• Citation:

@article{DBLP:journals/corr/abs-2104-07091,
  author    = {Mingda Chen and
               Zewei Chu and
               Sam Wiseman and
               Kevin Gimpel},
  title     = {SummScreen: {A} Dataset for Abstractive Screenplay Summarization},
  journal   = {CoRR},
  volume    = {abs/2104.07091},
  year      = {2021},
  url       = {https://arxiv.org/abs/2104.07091},
  archivePrefix = {arXiv},
  eprint    = {2104.07091},
  timestamp = {Mon, 19 Apr 2021 16:45:47 +0200},
  biburl    = {https://dblp.org/rec/journals/corr/abs-2104-07091.bib},
  bibsource = {dblp computer science bibliography, https://dblp.org}
}

summscreen/fd (default config)

• Config description: ForeverDreaming

• Dataset size: 132.99 MiB

• Auto-cached (documentation): Yes

• Splits:

Split	Examples
'test'	337
'train'	3,673
'validation'	338

• Feature structure:

FeaturesDict({
    'episode_number': Text(shape=(), dtype=string),
    'episode_title': Text(shape=(), dtype=string),
    'recap': Text(shape=(), dtype=string),
    'show_title': Text(shape=(), dtype=string),
    'transcript': Text(shape=(), dtype=string),
    'transcript_author': Text(shape=(), dtype=string),
})

• Feature documentation:

Feature	Class	Shape	Dtype	Description
	FeaturesDict
episode_number	Text		string
episode_title	Text		string
recap	Text		string
show_title	Text		string
transcript	Text		string
transcript_author	Text		string

• Examples (tfds.as_dataframe):

summscreen/tms

• Config description: TVMegaSite

• Dataset size: 592.53 MiB

• Auto-cached (documentation): No

• Splits:

Split	Examples
'test'	1,793
'train'	18,915
'validation'	1,795

• Feature structure:

FeaturesDict({
    'episode_summary': Text(shape=(), dtype=string),
    'recap': Text(shape=(), dtype=string),
    'recap_author': Text(shape=(), dtype=string),
    'show_title': Text(shape=(), dtype=string),
    'transcript': Text(shape=(), dtype=string),
    'transcript_author': Tensor(shape=(None,), dtype=string),
})

• Feature documentation:

Feature	Class	Shape	Dtype	Description
	FeaturesDict
episode_summary	Text		string
recap	Text		string
recap_author	Text		string
show_title	Text		string
transcript	Text		string
transcript_author	Tensor	(None,)	string

• Examples (tfds.as_dataframe):