README.md

metadata

license: cc-by-nc-sa-4.0

Dataset Description

The dataset is collected and used in paper Modeling Empathetic Alignment in Conversation. It contains therapeutic Reddit conversations labeled with 9,284 appraisals from both the Target and Observer and 3,262 alignments between the Target and Observer.

The dataset has the following entries:

id	target_id	observer_id	parent_id	subreddit	target_text	observer_text	distress_score	condolence_score	empathy_score	full_text	spans	alignments
int	string	string	string	string	string	string	float	float	float	string	[start, end, label], ... ]	[[(target_start, target_end), (observer_start, observer_end)], ...]

• id: the id of current instance
• target_id (our project specific): the id of target text from Reddit API
• observer_id (our project specific): the id of observer text from Reddit API
• parent_id (our project specific): the parent_id of comment or post from Reddit API
• subreddit (our project specific): Subreddit name
• target_text: target text
• observer_text: observer text
• distress_score: the distress score for target (refer to the paper for getting the score)
• condolence_score: the condolence score for observer (refer to the paper for getting the score)
• empathy_score: the empathy score for target-observer pair (refer to the paper for getting the score)
• full_text: the combination of target_text and observer_text with the format of target:\n\n${target_text}\n\nobserver:\n\n${observer_text}
• spans: annotated appraisals, including Pleasantness, Anticipated Effort, Certainty, Objective Experience, Self-Other Agency, Situational Control, Advice, Trope
• alignments: annotated alignments

Data Sources

Candidate data for annotation was selected from all post-comment pairs and comment-reply to those posts in 35 English-language Subreddits from 2019-01 to 2021-06.

Annotation Guidelines

Please refer to codebook for more details.