slupart/ikat23-rewrite-mistral

iKAT 2023 Rewritten Queries (Mistral-based Rewrites from DiSCo)

This dataset contains Mistral-generated rewrites of conversational queries from the TREC iKAT 2023 dataset, used for open-domain conversational question answering.

It provides:

• One rewritten query per turn, generated using the Mistral-7B-Instruct-v0.2 model
• One split (test) aligned with the original iKAT 2023 data

These rewrites were used in the paper:

• DiSCo SPLADE - SIGIR 2025 full paper by Lupart et al.: https://arxiv.org/abs/2410.14609

For more information on the rewriting method and evaluation setup, see the DiSCo paper (Lupart et al., 2024).

---

Dataset Structure

The dataset consists of entries with:

• id (string): Turn identifier in the format Conversation_no_Turn_no (e.g., 12_1)
• text (string): A rewritten query generated by the Mistral model

Usage

from datasets import load_dataset

# Load the dataset from the Hugging Face Hub
dataset = load_dataset("ikat23-rewrite-mistral")

# Show the first 5 examples from the test split
print("\nTest Split:")
for example in dataset["test"].select(range(5)):
    print(example)

Citation

If you use our dataset, please cite our work and the original TREC iKAT 2023 dataset:

@article{lupart2024disco,
  title={DiSCo Meets LLMs: A Unified Approach for Sparse Retrieval and Contextual Distillation in Conversational Search},
  author={Lupart, Simon and Aliannejadi, Mohammad and Kanoulas, Evangelos},
  journal={arXiv preprint arXiv:2410.14609},
  year={2024}
}

@article{aliannejadi2024trec,
  title={Trec ikat 2023: The interactive knowledge assistance track overview},
  author={Aliannejadi, Mohammad and Abbasiantaeb, Zahra and Chatterjee, Shubham and Dalton, Jeffery and Azzopardi, Leif},
  journal={arXiv preprint arXiv:2401.01330},
  year={2024}
}