uss-ratings-dataset / README.md
akomma's picture
Update README.md
0d89e2f
metadata
license: mit
task_categories:
  - text-classification
  - zero-shot-classification
  - conversational
language:
  - en
pretty_name: uss-ratings-dataset
size_categories:
  - 10K<n<100K

Dataset Description

Dataset Summary

  • Dialogs Quality Dataset
  • With both turn-level and dialog-level ratings provided on a scale of 1 to 5 by human annotators.
  • Each task has been annotated by multiple annotators.
  • Contains annotated dialogs from 4 different datasets (SGD, MultiWoz, ReDial, CCPE)
  • Total 34358 turns from 3500 dialogs
Dataset Dialogs Turns
SGD 1000 11833
MWOZ 1000 10553
Redial 1000 6792
CCPE 500 5180

Column Definitions

Column Type Example Value Description
split str CCPE;MWOZ;SGD;Redial dataset name
session_idx int 1 dialog identifier
turn_idx int 1 turn identifier within a dialog
tree_idx int 1 tree identifier within a turn (is all 1s here)
system str Do you like movies system message
user str No I don't like user message
turn_scores list [3; 2; 2] list of turn-level quality scores from different human annotations
mean_turn_rating float 2.33 mean of turn-level annotator scores
mode_turn_rating int 2 mode of turn-level annotator scores
dialog_scores list [3; 3; 3] list of dialog-level quality scores from different human annotations
mean_dialog_rating float 3.00 mean of dialog-level annotator scores
mode_dialog_rating int 3 mode of dialog-level annotator scores

Dataset Description