README.md

---
dataset_info:
  features:
  - name: personas
    sequence: string
  - name: additional_context
    dtype: string
  - name: previous_utterance
    sequence: string
  - name: context
    dtype: string
  - name: free_messages
    sequence: string
  - name: guided_messages
    sequence: string
  - name: suggestions
    struct:
    - name: convai2
      sequence: string
    - name: empathetic_dialogues
      sequence: string
    - name: wizard_of_wikipedia
      sequence: string
  - name: guided_chosen_suggestions
    sequence: string
  splits:
  - name: train
    num_bytes: 8990720
    num_examples: 4096
  - name: validation
    num_bytes: 1601536
    num_examples: 723
  download_size: 10600448
  dataset_size: 10592256
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train.parquet
  - split: validation
    path: data/validation.parquet
size_categories:
- 1K<n<10K
license: apache-2.0
task_categories:
- text-generation
---

## Compatibility Update

This repository is a compatibility-fixed version of the original **Blended Skill Talk** dataset.

The original dataset can be found at:

- Original Hugging Face dataset: https://huggingface.co/datasets/anezatra/blended-skill-talk

This version was created to maintain compatibility with newer versions of the Hugging Face `datasets` library.

### Changes from the Original Dataset

The following changes were made:

- Removed the unused `label_candidates` column.
- Removed the invalid `label_candidates` feature definition (`sequence: null`) that caused schema parsing failures.
- Re-exported the dataset Parquet files with a corrected schema.
- Updated dataset metadata to match the exported files.

No conversation examples were modified.

No examples were added or removed.

The dataset content remains the same as the original release.

---

# Blended Skill Talk

## Dataset Summary

Blended Skill Talk is a conversational dataset designed to train and evaluate dialogue systems capable of combining multiple conversational skills.

The dataset combines persona-based conversation, empathetic dialogue, and knowledge-grounded conversation elements to encourage models to produce engaging, context-aware responses.

The dataset incorporates information from ConvAI2, EmpatheticDialogues, and Wizard of Wikipedia.

---

## Data Structure (Updated)

### Fields

| Field | Description |
|:------|:-------------|
| personas | List of personas participating in the conversation |
| additional_context | Extra context or scenario description |
| previous_utterance | The immediately preceding dialogue turn(s) |
| context | General conversation context |
| free_messages | Free-form user or system messages |
| guided_messages | Messages generated via guided prompts |
| suggestions | Suggested responses from ConvAI2 Empathetic Dialogues and Wizard of Wikipedia |
| guided_chosen_suggestions | Selected guided suggestions associated with the conversation |

---

### Splits

| Split | Examples | Size (bytes) | Description |
|:------|----------|--------------|-------------|
| Train | 4096 | 8990720 | Used for model training |
| Validation | 723 | 1601536 | Used for validation and tuning |

**Total dataset size:** 10592256 bytes  
**Total number of dialogues:** 4819

---

## Loading This Version

```python
from datasets import load_dataset

dataset = load_dataset(
    "TutorialGuide/blended-skill-talk-fixed"
)

print(dataset)
```

---

## References

Original Hugging Face dataset: https://huggingface.co/datasets/anezatra/blended-skill-talk

Smith, E. M., Williamson, M., Shuster, K., Weston, J., & Boureau, Y. L. (2020). Can You Put it All Together: Evaluating Conversational Agents' Ability to Blend Skills. *arXiv preprint arXiv:2004.08449*. ([https://arxiv.org/abs/2004.08449](https://arxiv.org/abs/2004.08449))