This dataset was created with the goal of identifying direct reply relationships, and it therefore labels direct reply relationships between pairs of messages in a transcript.

The points below explain the format of the data.

• message.json is a JSON representation of chat transcript available to the annotators.
• The remaining three files (e.g., annot1.json) contain the annotations made by each user.
• Each line in the file is a particular save state, and the final line contains the final state of the annotation.
• The state of the annotation is a list, with the i-th index correponding to a list of parents for the i-th message.

To better understand the annotation procedure for this dataset, the annotation software has been open-sourced.

Contact me if you have any questions about the creation or usage of this dataset.