Where Should I Study? Biased Language Models Decide! Evaluating Fairness in LMs for Academic Recommendations

TL;DR

This work analyzes biases in three open‑source LLMs used for academic recommendations by evaluating 360 synthetic profiles against over 25,000 recommendations. It introduces two fairness metrics, Demographic Representation Score (DRS) and Geographic Representation Score (GRS), to quantify demographic fit and geographic diversity, with definitions such as $DRS = w_1 Acc + w_2 Rep + w_3 Acad$ and $GRS(c)=\sqrt{Scaled\_Repr(c) \cdot Rep\_covg(c)}$. The results reveal persistent Western‑centric bias, gender stereotypes, and socio‑economic stratification across models, even though Llama‑3.1 shows comparatively broader coverage. The study provides a reproducible evaluation framework and urges bias mitigation in educational LLMs to improve global access to higher education.

Abstract

Large Language Models (LLMs) are increasingly used as daily recommendation systems for tasks like education planning, yet their recommendations risk perpetuating societal biases. This paper empirically examines geographic, demographic, and economic biases in university and program suggestions from three open-source LLMs: LLaMA-3.1-8B, Gemma-7B, and Mistral-7B. Using 360 simulated user profiles varying by gender, nationality, and economic status, we analyze over 25,000 recommendations. Results show strong biases: institutions in the Global North are disproportionately favored, recommendations often reinforce gender stereotypes, and institutional repetition is prevalent. While LLaMA-3.1 achieves the highest diversity, recommending 481 unique universities across 58 countries, systemic disparities persist. To quantify these issues, we propose a novel, multi-dimensional evaluation framework that goes beyond accuracy by measuring demographic and geographic representation. Our findings highlight the urgent need for bias consideration in educational LMs to ensure equitable global access to higher education.

Figures (19)

• Figure 1: Demographic and geographic biases in university and program recommendations. X and Y represent controlled demographic placeholders in this setup.
• Figure 2: Overview of the key perspectives and components of the evaluation framework
• Figure 3: Prompt template and augmentation setup used for university recommendation experiments
• Figure 4: Program Recommendation trends by gender under the base prompt and the context-augmented prompt (A) with engineering background.
• Figure 5: Distribution of the top 20 most frequently recommended university locations across the three models (Mistral, LLaMA, and Gemma).