README.md

---
language:
- en
license: apache-2.0
size_categories:
- 1K<n<10K
task_categories:
- text-generation
- text-classification
dataset_info:
  features:
  - name: experience_type
    dtype: string
  - name: emotion_type
    dtype: string
  - name: problem_type
    dtype: string
  - name: situation
    dtype: string
  - name: survey_score
    struct:
    - name: seeker
      struct:
      - name: empathy
        dtype: string
      - name: final_emotion_intensity
        dtype: string
      - name: initial_emotion_intensity
        dtype: string
      - name: relevance
        dtype: string
    - name: supporter
      struct:
      - name: relevance
        dtype: string
  - name: seeker_question1
    dtype: string
  - name: seeker_question2
    dtype: string
  - name: supporter_question1
    dtype: string
  - name: supporter_question2
    dtype: string
  - name: conversations
    list:
    - name: content
      dtype: string
    - name: role
      dtype: string
    - name: strategy
      dtype: string
  splits:
  - name: train
    num_bytes: 3089088
    num_examples: 910
  - name: test
    num_bytes: 668538
    num_examples: 195
  - name: valid
    num_bytes: 663512
    num_examples: 195
  download_size: 2191251
  dataset_size: 4421138
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: test
    path: data/test-*
  - split: valid
    path: data/valid-*
tags:
- esconv
- empathetic
---
# ESCONV for LLM
 This repository contains a reformatted version of the ESCONV dataset, tailored for seamless integration with Language Model (LLM) training and inference. The original dataset's format posed challenges for direct application in LLM tasks, prompting us to restructure and clean the data. 

## Data Restructuring
1. Assigned the `user` role to the `usr`, `assistant` role to the `sys`.
2. Removed the `survey_scor` and 'supporter' fields to streamline the data.

## Data Format
Each entry in the reformatted dataset consists of the following fields: 

- conversations: A list of dictionaries, where each dictionary represents a turn in the dialogue and contains:
  - role: A string indicating the speaker's role, either user or assistant.
  - content: A string containing the dialogue content.
  - strategy: A string containing the strategy of current dialogue content, if role is user, strategy is NONE.
- emotion: A string indicating the emotional label associated with the dialogue (corresponds to the emotion_type field in the original dataset).
- situation: A string describing the situational label for the dialogue (corresponds to the situation field in the original dataset).
- problem: A string describing the problem label for the user (corresponds to the problem_type field in the original dataset).
- experience: A string, corresponds to the experience_type field in the original dataset.


## Dataset Statistics

| Dataset     | Total Turn | Average Turn | Average Length |
|-------------|------------|--------------|----------------|
| Train       | 26,648     | 29.284       | 14.547         |
| Validation  | 5,678      | 29.118       | 14.630         |
| Test        | 6,039      | 30.969       | 13.756         |