Exam Variants: Human-in-the-Loop AIG Quantitative Methods & Data Mining Spring 25
Description
Can instructors' efficiently create multiple parallel exams with an AI assistant while maintaining fairness and rigour? This is a collection of two dozen exam variants (instructions & answer keys) created through a human-in-the-loop AI-Instructor approach to automatic item generation. Files include Midterm and Final exam variants for two courses: BUS306 Quantitative Methods and BUS397 Data Mining.
Files
Steps to reproduce
Two optimized example prompts from the initial phase that can be used to generate similar exam variants are: Quantitative Methods “Generate exam variants that clearly instructs students to calculate stock volatility for verifiably listed stocks from the NYSE, NASDAQ, London, Hong Kong, or NIKKEI stock exchanges over a specific 30-day period (provide exact start and end dates, e.g., Sept 8, 2020 – Oct 8, 2020). Include guidance that volatility should be computed as the standard deviation of daily returns and ensure no ambiguity in what data or date range is required. Exams should test the following learning objectives: procedural fluency, emergent statistical thinking, and explanatory literacy; related to their analysis of the dataset provided. Avoid culturally specific references to key dates (e.g. Black Friday)” And Data Mining: “Generate an exam question that asks students to analyze Airbnb data across several cities (for example, Austin, Amsterdam, Cape Town, Istanbul). Focus on hosts owning between 2 and 5 properties. Before finalizing the question, double-check that using a threshold in this range yields at least 100 properties in each city’s dataset to allow meaningful analysis. The question should explicitly instruct students how to identify those properties and what metrics to compute. Students should be able to join, clean, analyze and interpret results using SQL.” Note: Full links to instructor chats with GPT o4 mini-high may be provided upon request (cburke@fus.edu).