Nattapong Sanchan Bangkok University

Gold Standard Online Debates Summaries

Salient Sentence Selection Dataset (SSSD)

About this dataset

This dataset consists of comments collected from 11 online debate topics. Each debate topic consists of two opposing sides: Agree (Yes) and Disagree (No). It was primarily used in the salient-sentence-selection process, one step in online debate summarization. Each comment was annotated by 5 judges, based on a 20% compression rate. Additionally, one sentence was annotated from each comment, as each comment was considered to express a salient piece of information. More detail on this corpus is in the paper below.

XML format

Online debate comments are stored in XML format, as shown in the figure below.

Example XML structure of the SSSD dataset

Each debate contains several comments (indicated by <comment id="">). Each comment is split into sentences (indicated by <sentence id="">). Each comment was annotated at a 20% compression rate, stored in the <annotation> tag. Other useful information collected includes the side of each comment (Agree or Disagree with the debate topic) and the number of likes supporting that comment.

CSV format

The dataset contains the following attributes:

  1. recordid: the record identification number that uniquely identifies each sentence.
  2. debateid: the identification number of each debate.
  3. debatetopicname: the name of a debate topic.
  4. commentid: the identification number of each debate comment.
  5. sentenceid: the identification number of a sentence in each debate comment.
  6. sentence: a sentence in each comment.
  7. side: the side (stance) of a debate comment.
  8. like: the number of votes supporting this comment.
  9. annotation: the annotation for a comment.

The comments in the CSV version are the same as those in the XML format; the difference is how they are organized, as shown below. The boundary for each comment is separated by the sentenceid column. For instance, commentid 10 is in debateid DTP03 and contains five sentences (rows 0–4); for this comment, 5 annotators manually selected sentenceid 2, 2, 2, 2, and 5 respectively.

Example CSV structure of the SSSD dataset

Download this dataset

If you use this dataset in your work, please cite the paper below. By downloading this dataset, you agree that the data will be used for educational purposes only.

Download XML format Download CSV format

References

Download this paper (PDF) Download BibTeX

BibTeX entry 
@inproceedings{2018_Sanchan_Aker_Bontcheva,
  author    = {Sanchan, Nattapong and Aker, Ahmet and Bontcheva, Kalina},
  title     = {Gold Standard Online Debates Summaries and First Experiments Towards Automatic Summarization of Online Debate Data},
  booktitle = {Computational Linguistics and Intelligent Text Processing},
  editor    = {Gelbukh, Alexander},
  year      = {2018},
  address   = {Cham},
  volume    = {10762},
  series    = {Lecture Notes in Computer Science},
  publisher = {Springer International Publishing},
  pages     = {495--505},
  isbn      = {978-3-319-77116-8}
}
Rich-text citation (copy & paste)

Sanchan, N., Aker, A., Bontcheva, K. (2018) Gold Standard Online Debates Summaries and First Experiments Towards Automatic Summarization of Online Debate Data. In: Gelbukh, A. (eds) Computational Linguistics and Intelligent Text Processing. CICLing 2017. Lecture Notes in Computer Science, vol 10762. Springer, Cham.