agentlans/knkarthick-dialogsum

DIALOGSum Corpus in ShareGPT-like format

DialogSum is a large-scale dialogue summarization dataset, now reformatted into a ShareGPT-like format for easier use in chat-based language models.

Dataset Description

This dataset is a modified version of the original DialogSum corpus, with the following key changes:

1. Dialogues have been reformatted into a ShareGPT-like structure.
2. A system prompt has been created from the summary, indicating the conversation scenario and the roles of #Person1# (the first speaker) and #Person2# (second speaker).
3. Only rows where #Person1# or #Person2# are mentioned in the summary and there are only two people in the conversation are included. This helps clarify who is playing which role in the chat.

The dataset was processed from the knkarthick/dialogsum dataset.

Dataset Statistics

The dataset contains a total of 12,586 entries, distributed as follows:

• Test set: 1 258 entries
• Train set: 10 883 entries
• Validation set: 445 entries

Data Format

Each entry in the dataset is a JSON object with the following structure:

{
  "id": "train_1825",
  "conversations": [
    {
      "from": "system",
      "value": "#Person1# and Margaret talk about the arrangements for their company retreat. Margaret wonders the criteria for staff to attend, and #Person1# says the invitations are for the management team.\n\nThe user is #Person1#. You are #Person2#."
    },
    {
      "from": "human",
      "value": "Margaret, I'd like to follow up with the arrangements for our company retreat, to take place next month. We need to book the venue, make arrangements for speakers and door prizes, and set up all the activities and accommodations. Also, very important, we need to determine which of our staff will be eligible and will be available to go. We've got to get a head count in order to make reservations."
    },
    {
      "from": "gpt",
      "value": "What's the criteria for staff to attend? Are we only including our management team in this affair? Or will we be extending invitations to lower level employees to attend?"
    }
  ],
  "summary": "#Person1# and Margaret talk about the arrangements for their company retreat. Margaret wonders the criteria for staff to attend, and #Person1# says the invitations are for the management team.",
  "topic": "company retreat arrangements"
}

Usage

This dataset can be used for training and evaluating chat-based language models, particularly those focused on dialogue summarization tasks. The ShareGPT-like format makes it suitable for fine-tuning large language models on conversational data.

License

The license for this dataset is inherited from the original DialogSum dataset. Please refer to the original dataset for licensing information[3].

Citation

If you use this dataset, please cite the original DialogSum paper:

@inproceedings{chen-etal-2021-dialogsum,
    title = "{D}ialog{S}um: {A} Real-Life Scenario Dialogue Summarization Dataset",
    author = "Chen, Yulong  and
      Liu, Yang  and
      Chen, Liang  and
      Zhang, Yue",
    booktitle = "Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021",
    month = aug,
    year = "2021",
    address = "Online",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2021.findings-acl.449",
    doi = "10.18653/v1/2021.findings-acl.449",
    pages = "5062--5074",
}

Acknowledgements

Thank to the creators of the original DialogSum dataset.

See Also

agentlans/allenai-soda for a similar dataset also in ShareGPT-like format.