Skip to content
Open access

Beyond Preference Matching: Exploration-Aware Calibration for Diverse Recommendations

Oct 2026 · Brazilian Symposium on Multimedia and the Web · 0 citations · 22 references

Abstract

Most calibration approaches in recommender systems focus on matching the distribution of user preferences observed in historical interactions. Although effective for improving preference representation, reproducing historical preference distributions may reinforce already concentrated user profiles, limiting content discovery and recommendation diversity. We propose Exploration-Aware Calibration (EAC), a calibration framework that explicitly incorporates category-level exploration into the recommendation process. Inspired by exploration-exploitation principles, EAC assigns exploration bonuses to underrepresented categories and integrates them into a calibration-based re-ranking strategy. We evaluate the proposed method using multiple recommendation algorithms in both movie and music domains. Results show that EAC consistently improves intra-list diversity, category coverage, and reduces recommendation concentration while preserving competitive levels of recommendation accuracy. Our findings demonstrate that introducing controlled exploration into calibration offers a practical and effective approach for balancing relevance and diversity in recommender systems.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.