Can we predict HW grades from ChatGPT interactions?
• Let’s make a unique and challenging project
• We don’t know if this is possible and we look at your approach to address this problem
• You all learn by grading other project teams and see their approaches.
Format of the project
• You will receive the following materials
• Collection of HTML files that contain ChatGPT histories (150-200 files)
• CSV file that has a corresponding score of the student
• Output expected
• Github repositories for all the material (code, scripts, etc.)
• README file that explains your approach clearly and your findings and contribution of team members.
Structure of the README file
• Overview of the repository: Link different script and code pieces used and explain their function for the project
• Methodology: High-level explanation of things considered and solutions offered.
• Results: Experimental findings supported by figures, tables etc.
• Team contributions: List all team members by their names and how they contributed to the project
Challenges
• Extracting information from ChatGPT
(feature engineering)
• Not all questions in the assignment solved using ChatGPT (missing data + imputation)
• Mapping prompt&answer to questions
• Dataset sizes
• Predicting scores (model building and evaluation)
• Clearly communicating your findings and methodologies used for the task
Evaluation
• Code and material review (30%)
• Quality of the project ideas and results (40%)
• Peer-evaluation
• Grading other projects (10%)
• Grades received from peers (20%)
What is the submission format
• We will only ask for a public Github repository URL.
Each team will submit “only one” repo.
• We will automate cloning Github repositories with a script.
Mistakes in the URL format will cause you points!
Deadlines
PLEASE PLAN ACCORDINGLY!
