FACTER: Fairness-Aware Conformal Thresholding and Prompt Engineering for Enabling Fair LLM-Based Recommender Systems

Link: http://arxiv.org/abs/2502.02966v1

PDF Link: http://arxiv.org/pdf/2502.02966v1

Summary: We propose FACTER, a fairness-aware framework for LLM-based recommendationsystems that integrates conformal prediction with dynamic prompt engineering.

By introducing an adaptive semantic variance threshold and aviolation-triggered mechanism, FACTER automatically tightens fairnessconstraints whenever biased patterns emerge.

We further develop an adversarialprompt generator that leverages historical violations to reduce repeateddemographic biases without retraining the LLM.

Empirical results on MovieLensand Amazon show that FACTER substantially reduces fairness violations (up to95.

5%) while maintaining strong recommendation accuracy, revealing semanticvariance as a potent proxy of bias.

Published on arXiv on: 2025-02-05T08:07:04Z