ProfessorBob/instruct-MultiQ3

Educational Question Generation Dataset

Overview

• Dataset Name: Educational Question Generation Dataset
• Dataset Identifier: v1.0
• Dataset Language: French
• Dataset Domain: Educational
• Dataset Size:
  • Training set: 5083 examples
  • Validation set: 176 examples
  • Test set: 188 examples
• Dataset Description: This dataset is designed to train a Language Model (LLM) on educational materials to generate useful questions for students. Each example includes an instruction and an input text, and the goal is to generate relevant questions based on the input text.

Intended Use

• Primary Task: Question generation for educational purposes
• Potential Use Cases:
  • Automated question generation for educational platforms
  • Supplemental resource creation for teachers
  • Question generation for tutoring systems

Dataset Composition

• Data Collection Process: The dataset was created by collecting educational texts and asking annotators to generate interesting questions based on those texts.
• Data Sources: Le livre scolaire and Wikipédia, we focused on History, Geography, and life science.
• Data Preprocessing: We cutted the last sentences of input longer than 1200 words.

Dataset Information

• Number of Examples:
  • Training set: 5083 examples
  • Validation set: 176 examples
  • Test set: 188 examples
• Dataset Structure: Each example consists of the following fields:
  • instruction: A fixed instruction to generate interesting questions for students based on the text.
  • input: The text on which the questions should be based.
  • output: The questions generated by annotators. (from 1 to 3 questions, if more we cut it to an other data with the same input text but different output)

Dataset Quality

• Data Quality Assurance: Annotators were provided with guidelines to ensure the quality and relevance of the generated questions.
• Limitations and Biases: The dataset's quality heavily relies on the expertise and subjectivity of the annotators. The generated questions may not cover all possible aspects or be suitable for every educational context.

Dataset Maintenance

• Dataset Curators: LISN / Stellia / SATT-Paris-Saclay
• Dataset Versioning: The current version of the dataset is v1.0.

Related Datasets

• Related Datasets: [Link to full qacr dataset]

Legal and Ethical Considerations

• License: Dataset request will not be accepted until we tackle this issue.
• Citation Information: Please use the following citation when referencing this dataset:

[Insert citation information]