Education Dialogue Dataset

Data Description

• Contains 40,000 training samples and 7,234 test samples.
• Each sample is a complete teacher‑student conversation, including topic and teacher/student preference metadata.

Data Format

• The data consist of six JSON files: five for training and one for testing.
• Each dialogue entry includes the following fields:
  • background_info: context information containing:
    • topic: the subject the teacher needs to teach.
    • student_prefrences: the student's preferred learning mode, e.g., lecture‑style or hands‑on activity.
    • teacher_prefrences: the teacher's preferred teaching mode, e.g., lecture‑style or hands‑on activity.
    • student_reactions: the student's reaction if not taught in their preferred way, e.g., loss of interest or possible adaptation.
    • teacher_reactions: the teacher's reaction if the student is not taught in the teacher's preferred way, e.g., feeling frustrated or possible adaptation.
  • conversation: a list of exchanges between teacher and student; each turn includes a role field identifying the speaker and a text field containing the utterance.

Data Generation

• Data were generated by prompting Gemini Ultra with the following instructions:
  • Simulate a learning scenario where ... (content omitted to preserve original prompting details).