Russian-Language Dialogues Dataset (RLDD)

Dataset Overview

This dataset contains approximately 9 million Russian-language dialogues collected from diverse sources, including:

Each dialogue is tagged with a generalized source identifier, enabling targeted filtering. The dataset has undergone basic preprocessing, including deduplication, cleaning, typographical normalization, and minor typo correction.

Sample dialogue:

- Вы уже кому-нибудь читали свои стихи?
- Нет, а что?
- Просто у Вас глаз подбит.

Each conversation turn is started with em dash symbol.

Intended Use

This dataset is designed for:

Statistical research in linguistics
Supplementary training data for pretraining small language models
Experiments in conversational AI and NLP tasks

Usage

from datasets import load_dataset

for sample in load_dataset("inkoziev/Conversations", split="train", trust_remote_code=True, streaming=True):
  print(sample['conversation'])

Citing

If you use this dataset in your research or projects, please cite it as follows:

@misc{Conversations,
  author       = {Ilya Koziev},
  title        = {Russian-Language Dialogues Dataset},
  year         = {2025},
  publisher    = {Hugging Face},
  howpublished = {\url{https://huggingface.co/datasets/inkoziev/Conversations}},
}

License

This dataset is licensed under the CC-BY-NC-4.0 license, which permits non-commercial use only. For commercial use, please contact the author at [[email protected]].

By using this dataset, you agree to:

Provide proper attribution to the author.
Refrain from using the dataset for commercial purposes without explicit permission.

Limitations

Automated Processing: The dataset was processed automatically with only selective manual checks. As a result, some dialogues may contain errors such as neighboring lines incorrectly split between participants.
Uncorrected Errors: Grammatical errors, typos, and spelling mistakes present in the original sources have been preserved. These may impact the performance of language models trained on this dataset.
Data Quality Variability: Due to the diverse sources of the dialogues, the quality and consistency of the data may vary significantly.

Ethical considerations

The dataset includes dialogues sourced from internet forums, which may contain:

Profanity and offensive language
Insults or derogatory remarks
Misinformation or fake facts

Users of this dataset should be aware of these potential issues and take appropriate measures to filter or handle such content responsibly, especially when deploying models trained on this data in real-world applications.