Hugging Face's logo

Datasets:
Samsung
/
samsum

Tasks:
Summarization
Languages:
English
Size:
10K<n<100K
ArXiv:
Tags:
conversations-summarization
License:
Dataset card Files Community
11
samsum / README.md
lhoestq's picture
lhoestq HF staff
add dataset_info in dataset metadata
c86cd37
preview code
|
raw
history blame
6.88 kB
metadata
annotations_creators:
  - expert-generated
language_creators:
  - expert-generated
language:
  - en
license:
  - cc-by-nc-nd-4.0
multilinguality:
  - monolingual
size_categories:
  - 10K<n<100K
source_datasets:
  - original
task_categories:
  - summarization
task_ids: []
paperswithcode_id: samsum-corpus
pretty_name: SAMSum Corpus
tags:
  - conversations-summarization
dataset_info:
  features:
    - name: id
      dtype: string
    - name: dialogue
      dtype: string
    - name: summary
      dtype: string
  config_name: samsum
  splits:
    - name: test
      num_bytes: 534492
      num_examples: 819
    - name: train
      num_bytes: 9479141
      num_examples: 14732
    - name: validation
      num_bytes: 516431
      num_examples: 818
  download_size: 2944100
  dataset_size: 10530064

Dataset Card for SAMSum Corpus

Table of Contents

Dataset Description

Dataset Summary

The SAMSum dataset contains about 16k messenger-like conversations with summaries. Conversations were created and written down by linguists fluent in English. Linguists were asked to create conversations similar to those they write on a daily basis, reflecting the proportion of topics of their real-life messenger convesations. The style and register are diversified - conversations could be informal, semi-formal or formal, they may contain slang words, emoticons and typos. Then, the conversations were annotated with summaries. It was assumed that summaries should be a concise brief of what people talked about in the conversation in third person. The SAMSum dataset was prepared by Samsung R&D Institute Poland and is distributed for research purposes (non-commercial licence: CC BY-NC-ND 4.0).

Supported Tasks and Leaderboards

[Needs More Information]

Languages

English

Dataset Structure

Data Instances

The created dataset is made of 16369 conversations distributed uniformly into 4 groups based on the number of utterances in con- versations: 3-6, 7-12, 13-18 and 19-30. Each utterance contains the name of the speaker. Most conversations consist of dialogues between two interlocutors (about 75% of all conversations), the rest is between three or more people

The first instance in the training set: {'id': '13818513', 'summary': 'Amanda baked cookies and will bring Jerry some tomorrow.', 'dialogue': "Amanda: I baked cookies. Do you want some?\r\nJerry: Sure!\r\nAmanda: I'll bring you tomorrow :-)"}

Data Fields

  • dialogue: text of dialogue.
  • summary: human written summary of the dialogue.
  • id: unique id of an example.

Data Splits

  • train: 14732
  • val: 818
  • test: 819

Dataset Creation

Curation Rationale

In paper:

In the first approach, we reviewed datasets from the following categories: chatbot dialogues, SMS corpora, IRC/chat data, movie dialogues, tweets, comments data (conversations formed by replies to comments), transcription of meetings, written discussions, phone dialogues and daily communication data. Unfortunately, they all differed in some respect from the conversations that are typ- ically written in messenger apps, e.g. they were too technical (IRC data), too long (comments data, transcription of meetings), lacked context (movie dialogues) or they were more of a spoken type, such as a dialogue between a petrol station assis- tant and a client buying petrol. As a consequence, we decided to create a chat dialogue dataset by constructing such conversa- tions that would epitomize the style of a messenger app.

Source Data

Initial Data Collection and Normalization

In paper:

We asked linguists to create conversations similar to those they write on a daily basis, reflecting the proportion of topics of their real-life messenger conversations. It includes chit-chats, gossiping about friends, arranging meetings, discussing politics, consulting university assignments with colleagues, etc. Therefore, this dataset does not contain any sensitive data or fragments of other corpora.

Who are the source language producers?

linguists

Annotations

Annotation process

In paper:

Each dialogue was created by one person. After collecting all of the conversations, we asked language experts to annotate them with summaries, assuming that they should (1) be rather short, (2) extract important pieces of information, (3) include names of interlocutors, (4) be written in the third person. Each dialogue contains only one ref- erence summary.

Who are the annotators?

language experts

Personal and Sensitive Information

None, see above: Initial Data Collection and Normalization

Considerations for Using the Data

Social Impact of Dataset

[Needs More Information]

Discussion of Biases

[Needs More Information]

Other Known Limitations

[Needs More Information]

Additional Information

Dataset Curators

[Needs More Information]

Licensing Information

non-commercial licence: CC BY-NC-ND 4.0

Citation Information 

@inproceedings{gliwa-etal-2019-samsum,
    title = "{SAMS}um Corpus: A Human-annotated Dialogue Dataset for Abstractive Summarization",
    author = "Gliwa, Bogdan  and
      Mochol, Iwona  and
      Biesek, Maciej  and
      Wawer, Aleksander",
    booktitle = "Proceedings of the 2nd Workshop on New Frontiers in Summarization",
    month = nov,
    year = "2019",
    address = "Hong Kong, China",
    publisher = "Association for Computational Linguistics",
    url = "https://www.aclweb.org/anthology/D19-5409",
    doi = "10.18653/v1/D19-5409",
    pages = "70--79"
}

Contributions

Thanks to @cccntu for adding this dataset.