---
dataset_info:
  features:
  - name: sentence1
    dtype: string
  - name: sentence2
    dtype: string
  - name: labels
    dtype: string
  - name: metadata
    dtype: string
  splits:
  - name: test
    num_bytes: 5412865
    num_examples: 2895
  - name: validation
    num_bytes: 5087404
    num_examples: 2722
  - name: train
    num_bytes: 24100429
    num_examples: 12759
  download_size: 5704362
  dataset_size: 34600698
configs:
- config_name: default
  data_files:
  - split: test
    path: data/test-*
  - split: validation
    path: data/validation-*
  - split: train
    path: data/train-*
license: cc-by-nc-4.0
source_datasets:
- thu-coai/esconv
---

# esconv

ESConv supporter-strategy judgments from the authors' native dialogue splits.

Read the official JSON-lines files despite their .txt suffix and preserve
each annotated supporter turn separately, including consecutive turns with
different strategies. Some existing conversions merge those turns and lose
strategy alignment. Inputs contain only the situation, earlier dialogue and
current reply; surveys, strategy names and future turns are excluded.
Native conversations stay together; exact conversation duplicates favor
test, validation, then train. Original terms: CC BY-NC 4.0.

Original data: [thu-coai/esconv](https://huggingface.co/datasets/thu-coai/esconv), [original distribution](https://huggingface.co/datasets/thu-coai/esconv). Repackaged as parquet for [tasksource](https://github.com/sileod/tasksource) by [scripts/repackage_dataset/dialogue.py](https://github.com/sileod/tasksource/blob/main/scripts/repackage_dataset/dialogue.py#L110).

[Conversion provenance](provenance.json): source revisions, hashes, configuration and exclusions.