data-generator

input_data

• audio_transcripts
  • Transcripts of the ECE 120 course
• gpt-3
  • Question-Answer pairs generated by GPT-3
• patel_textbook_split
  • The patel textbook split into 200 words with 50 overlapping words on each side
• split_textbook
  • The ECE 120 textbook split into sentences, paragraphs, and sections

raw_data

• notes
  • The ECE 120 textbook
• patel_textbook
  • The Patel textbook

HF

Models used from HuggingFace.co

• huggingface.ipynb
  • end2end.json - ThomasSimonini/t5-end2end-question-generation
  • fine_tuned_data.json - mrm8488/t5-base-finetuned-question-generation-ap
  • Dataset Quality - Not Good

bart

Inspired from this paper's second best model using bart question generator and bert-mrc

• bart.ipynb
• Two different bert-mrc
  • bespin.json
  • ainze.json
  • Dataset Quality - Not Good

filtering

• cross_encoder.ipynb
  • Uses this model on HuggingFace.
• scoring.ipynb
  • Uses bleu scores
• Dialogue_RPT_Scoring.ipynb
  • Uses Dialogue RPT to rate the answers based on the Context of the GPT answer file