[Dataset] Emotional Support Conversation Dataset #315
Description
Basic info about the dataset
| Total Conversations | Total Speakers | Total Utterances |
|---|---|---|
| 1300 | 2600 | 38365 |
The Emotional-Support-Conversation dataset contains approximately 1,300 conversations between supporters and seekers. It is organized in a relatively clean format: the entire dataset is stored in a single JSON file, with the main conversations located under the "dialog" key. Each conversation typically consists of about 20 utterances. Each utterance includes the speaker label, researcher-provided annotations, and the actual textual content.
This is a ConvoKit-formatted version of the dataset originally distributed with the following paper:
https://arxiv.org/abs/2106.01144. Link to the original dataset https://github.com/thu-coai/Emotional-Support-Conversation
Dataset Details
Speaker-Level Information
- id : identifier given to a speaker, it is given in a format of {role}_{i}, where role is either "seeker" or "supporter", and i stands for the i-th they belongs
- metadata:
- role : could be either supporter or seeker
Utterance-Level Information
- conversation_id: the ID of the conversation this utterance belongs to
- reply_to: the ID of the previous utterance its reply to (None if it's the first utterance)
- speaker: speaker that initiated the utterance
- time_stamp: null for the entirety of this corpus. Did not include in the dataset
- text: content/text of the utterance
Conversation-Level Information
- metadata:
- experience_type: category describing the kind of life experience the seeker is going through
- emotion_type: emotion expressed by the seeker in the conversation
- problem_type: within the categories of ongoing depression | breakup with partner | job crisis | problems with friends | academic pressure | procrastination | alcohol abuse| issues with parent | sleep problems | appearance anxiety | school bullying | issues with children |
- situation: open text describing the causes of the emotional problem.
- survey_score: numeric score from post-conversation surveys assessing the quality of support received for both seeker and supporter. For example
'seeker': {'initial_emotion_intensity': '3', 'empathy': '3', 'relevance': '3', 'final_emotion_intensity': '2'}, 'supporter': {'relevance': '3'} - seeker_question1: reflection questions answered by the seeker about their experience during the conversation
- seeker_question2: additional reflection questions answered by the seeker about their experience during the conversation
- supporter_question1: reflection questions answered by the supporter about their experience during the conversation
- supporter_question2: additional reflection questions answered by the supporter about their experience during the conversation