Skip to content

MTS-Dialog

Bases: DatasetBuilder

MTS-Dialog (Medical Training Summarization Dialog) is a comprehensive dataset featuring 1.7k doctor-patient conversations, along with their corresponding summaries, including section headers and contents.

The dataset is structured as follows:

  • Training set: Comprises 1,201 pairs of conversations and associated summaries, aimed at facilitating the training of models for medical dialogue understanding and summarization.

  • Validation set: Contains 100 pairs of conversations and their summaries, used for model tuning and intermediate evaluation.

  • Test sets: Includes two distinct test sets, each with 200 conversations and corresponding section headers and contents:

    1. MTS-Dialog-TestSet-1-MEDIQA-Chat-2023.csv: Serves as the official test set for the MEDIQA-Chat 2023 challenge (Task A), focusing on chat-based medical consultations.

    2. MTS-Dialog-TestSet-2-MEDIQA-Sum-2023.csv: Used as the official test set for the MEDIQA-Sum 2023 challenge (Task A & Task B), emphasizing the summary generation from medical dialogues.

Paper: "MTS-Dialog: A New Dataset for Medical Training Summarization in Doctor-Patient Conversations" - https://aclanthology.org/2023.eacl-main.1681

Authors: Asma Ben Abacha, Wen-wai Yim, Yadan Fan, Thomas Lin

Dataset version from the GitHub repository: https://github.com/abachaa/MTS-Dialog

Source code in medplexity/benchmarks/mts_dialog/mts_dialog_dataset_builder.py 
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
class MTSDialogDatasetBuilder(DatasetBuilder):
    """
    MTS-Dialog (Medical Training Summarization Dialog) is a comprehensive dataset featuring 1.7k doctor-patient conversations, along with their corresponding summaries, including section headers and contents.

    The dataset is structured as follows:

    - Training set: Comprises 1,201 pairs of conversations and associated summaries, aimed at facilitating the training of models for medical dialogue understanding and summarization.

    - Validation set: Contains 100 pairs of conversations and their summaries, used for model tuning and intermediate evaluation.

    - Test sets: Includes two distinct test sets, each with 200 conversations and corresponding section headers and contents:

        1. MTS-Dialog-TestSet-1-MEDIQA-Chat-2023.csv: Serves as the official test set for the MEDIQA-Chat 2023 challenge (Task A), focusing on chat-based medical consultations.

        2. MTS-Dialog-TestSet-2-MEDIQA-Sum-2023.csv: Used as the official test set for the MEDIQA-Sum 2023 challenge (Task A & Task B), emphasizing the summary generation from medical dialogues.

    Paper: "MTS-Dialog: A New Dataset for Medical Training Summarization in Doctor-Patient Conversations" - <https://aclanthology.org/2023.eacl-main.1681>

    Authors: Asma Ben Abacha, Wen-wai Yim, Yadan Fan, Thomas Lin

    Dataset version from the GitHub repository: <https://github.com/abachaa/MTS-Dialog>
    """

    def __init__(self, loader: Loader = None):
        if loader is None:
            loader = MTSDialogGithubDatasetLoader()

        super().__init__(loader)

    def build_dataset(
        self,
        split_type: MTSDialogDatasetSplitType = "test",
        config=None,
    ) -> Dataset[MTSDialogDataPoint]:
        dialog_raw_data = self.loader.load(split_type)

        dialog_entries = [
            MTSDialogEntry(**dialog_raw) for dialog_raw in dialog_raw_data
        ]

        data_points = [
            MTSDialogDataPoint(
                id=str(dialog_entry.ID),
                input=MTSDialogInput(
                    dialog=dialog_entry.dialogue,
                ),
                expected_output=None,
                metadata=MTSDialogMetadata(
                    section_header=dialog_entry.section_header,
                    reference_summary=dialog_entry.section_text,
                ),
            )
            for dialog_entry in dialog_entries
        ]

        return Dataset[MTSDialogDataPoint](
            data_points=data_points, description=self.__doc__
        )