ZennyKenny/yandex-alice-sessions-large-syn

Synthetic Alice Dialog Sessions (Realistic, Large)

Alice is a GPT assistant created by Yandex and available for free use here.

Dataset Characteristics

Key Features

• session_id: Unique identifier for each conversation session
• user_id: Unique identifier for each simulated user
• device: Type of device used (smartphone, tv, smart_speaker, car_display, robot)
• timestamp: Simulated timestamp of each interaction
• role: Whether the speaker is "user" or "assistant"
• type: Response type (text, image, audio, video)
• content: The actual message content

Interaction Types

The dataset contains 5 main types of user requests:

1. Timers/Alarms: "Поставь таймер на 5 минут", "Поставь будильник на 7 утра"
2. Weather queries: "Какая погода?"
3. Media requests: "Покажи фото котов", "Включи музыку", "Включи радио"
4. TV controls: "Включи канал Россия 1", "Что по телеку?", "Какие новости?"
5. Navigation: "Построй маршрут до дома", "Как доехать до центра?"

Device Distribution

• Smartphone: 35% of sessions
• Smart Speaker: 30% of sessions
• TV: 20% of sessions
• Car Display: 10% of sessions
• Robot: 5% of sessions

Intended Use

This dataset is designed for:

• Training and evaluating voice assistants
• Studying human-AI interaction patterns
• Multimodal response generation research
• Device-specific interaction analysis

Limitations

1. Synthetic Nature: While designed to be realistic, all interactions are AI-generated simulations
2. Limited Diversity: Focuses on common assistant tasks rather than edge cases
3. Language: Only contains Russian language interactions
4. Temporal Patterns: Timestamps are simulated and don't reflect real usage patterns

Ethical Considerations

• Contains no real user data - all interactions are synthetic
• May contain biases present in the GPT-4o training data
• Simulated users represent a limited demographic range

Citation

If you use this dataset in research, please cite it as:

@misc{llmUsageRU2025,
  title = {Yandex Alice Sessions: Synthetic, Realistic, Large},
  author = {Hamilton, Kenneth},
  year = {2025},
  url = {https://huggingface.co/datasets/YourUsername/llm_usage_ru_dataset},
  note = {Synthetic dataset for LLM performance analysis in Russian.}
}