Jarbas/core_intents

Dataset: Core Intents

The Core Intents dataset contains feedback-related utterances classified into seven key categories. This dataset is designed to help train voice assistants in handling fundamental types of feedback and adjusting their behavior accordingly.

Labels

The dataset includes the following labels:

• ask_clarify: Requests for clarification, where the user seeks clearer explanations.
• ask_confirmation: Requests for confirmation, where the user asks to validate a previous command or statement.
• ask_repeat: Requests for repetition, where the user asks for something to be repeated.
• negative_feedback: Negative responses or corrections, indicating dissatisfaction or disagreement.
• positive_feedback: Positive responses, expressing satisfaction or agreement.
• neutral_feedback: Neutral responses, where the user shows indifference or general acceptance.
• stop: Commands indicating the user wants to halt an action or process.

Examples

Sample utterances for each label:

• ask_clarify:

  • "your response was not clear"
  • "your words were not so clear to me"

• ask_confirmation:

  • "can you check and confirm my last command please"
  • "can you check and confirm it"

• ask_repeat:

  • "would you try again please"
  • "would you try what you said again"

• negative_feedback:

  • "command wrong"
  • "dammit, I did not say it"

• positive_feedback:

  • "thanks, that's amazing"
  • "that is excellent, much appreciated"

• neutral_feedback:

  • "anything is fine"
  • "any one would be good to me"

Purpose

This dataset aims to cover core interactions a voice assistant should manage to provide responsive and adaptive behavior in real-time communication. It can be used in tasks such as:

• Intent classification
• Dialogue system training
• Feedback management in voice assistants