Datasets:

Languages:
English
ArXiv:
Tags:
License:

Train-dev-test splits ?

#1
by lihVerma - opened

This repo shows that data corresponds to only train split, but this dataset has all the dialogues corresponding to train-dev-test splits. Just check the github repo of this paper, where they have a process.py file, that mention, how to segregate it into respective splits and give you {train/dev/test}.txt files

Conversational AI (CoAI) group from Tsinghua University org

Thanks for your kind reminder. I just uploaded the data splits

lihVerma changed discussion title from Train-dev-test splits are already here to Train-dev-test splits ?

But in recent update to dataset, you have forgot to mention the Survey Score details. Quoting from the paper:-
Survey_Score details in ESConv dataset

Example

"survey_score": {
"seeker": {
"initial_emotion_intensity": "5",
"empathy": "5",
"relevance": "5",
"final_emotion_intensity": "1"
},
"supporter": {
"relevance": "5"
}
}

  • Supporter:

Post-chat Survey After each conversation, the supporter is asked to rate the extent that the seeker goes into detail about their problems on five-point Likert scales.

  • Seeker:

After each conversation, the help-seeker is asked to rate their emotion and the performance of the supporter on the following five-point Likert scales:

(1) Their emotion intensity after the emotional support conversation (a decrease from the intensity before the conversation reflects emotion improvement)

(2) the supporter’s empathy and understanding of the help-seeker’s experiences and feelings

(3) the relevance of the supporter’s responses to the conversation topic

Conversational AI (CoAI) group from Tsinghua University org

Oh yes. These meta data can be easily added by modifying process.py. I think I may give an update in case others need them.

Sign up or log in to comment