[Paper Review] Equality of Learning Opportunity via Individual Fairness in Personalized Recommendations

Mirko Marras, Ludovico Boratto+2 more|arXiv (Cornell University)|Jun 7, 2020

Online Learning and AnalyticsComputer Science105 references47 citations

TL;DR

This paper proposes a fairness-aware post-processing approach for personalized course recommendations that balances individual personalization with equality of learning opportunities. By formalizing seven educational principles—familiarity, validity, learnability, variety, quality, manageability, and affordability—into a composite fairness metric, the method reduces systemic inequalities in course recommendations across learners while incurring only a negligible drop in personalization performance on a real-world dataset of 40K+ learners and 30K courses.

ABSTRACT

Online educational platforms are playing a primary role in mediating the success of individuals' careers. Therefore, while building overlying content recommendation services, it becomes essential to guarantee that learners are provided with equal recommended learning opportunities, according to the platform values, context, and pedagogy. Though the importance of ensuring equality of learning opportunities has been well investigated in traditional institutions, how this equality can be operationalized in online learning ecosystems through recommender systems is still under-explored. In this paper, we formalize educational principles that model recommendations' learning properties, and a novel fairness metric that combines them in order to monitor the equality of recommended learning opportunities among learners. Then, we envision a scenario wherein an educational platform should be arranged in such a way that the generated recommendations meet each principle to a certain degree for all learners, constrained to their individual preferences. Under this view, we explore the learning opportunities provided by recommender systems in a large-scale course platform, uncovering systematic inequalities. To reduce this effect, we propose a novel post-processing approach that balances personalization and equality of recommended opportunities. Experiments show that our approach leads to higher equality, with a negligible loss in personalization. Our study moves a step forward in operationalizing the ethics of human learning in recommendations, a core unit of intelligent educational systems.

Motivation & Objective

• To operationalize equality of learning opportunities in online educational recommender systems.
• To identify and quantify systematic inequalities in course recommendations across learners.
• To develop a post-processing approach that enhances fairness without significantly compromising personalization.
• To formalize a fairness metric combining pedagogical principles to monitor and improve equity in recommendations.

Proposed method

• Defines seven educational principles—familiarity, validity, learnability, variety, quality, manageability, and affordability—as measurable properties of recommended courses.
• Constructs a composite fairness metric by combining these principles into a single score per learner using normalized, inverse-based formulations.
• Applies a greedy optimization algorithm to post-process recommendation lists, maximizing fairness while preserving personalization.
• Uses Hellinger distance for familiarity, Gini index for learnability, and normalized linear scaling for quality, manageability, and affordability.
• Employs a convex combination of personalization (via user feedback scores) and consistency with platform principles to guide the optimization.
• Proves the submodularity of the objective function, ensuring a (1−1/e)-approximation guarantee for the greedy optimization.

Experimental results

Research questions

• RQ1 To what extent do existing recommender systems in online education platforms produce unequal learning opportunities across learners?
• RQ2 How can fairness in course recommendations be formally measured by combining pedagogical principles such as quality, validity, and affordability?
• RQ3 Can a post-processing method improve fairness in recommendations without significantly degrading personalization?
• RQ4 What trade-offs exist between fairness and personalization in large-scale educational recommender systems?

Key findings

• The proposed post-processing approach achieves significantly higher equality of recommended learning opportunities across learners, with a measurable improvement in fairness metrics.
• The method incurs only a negligible loss in personalization performance, as evidenced by minimal degradation in user feedback-based ranking scores.
• Systematic inequalities were observed in baseline recommendations: learners received disproportionate access to high-quality, up-to-date, and affordable courses based on their interaction history.
• The fairness metric effectively captures disparities in course properties such as freshness, quality, and cost, revealing biases in standard collaborative filtering algorithms.
• The greedy optimization approach provides a near-optimal solution with a (1−1/e) approximation guarantee, ensuring computational efficiency.
• Empirical evaluation on a real-world dataset of 40,000+ learners and 30,000 courses confirms the method’s effectiveness in reducing fairness gaps while maintaining strong personalization.