<![CDATA[The automation of curriculum vitae (CV) classification raises major challenges related to accuracy, fairness, and the heterogeneity of candidate documents. Existing approaches often address these dimensions separately and struggle to reduce demographic bias while maintaining high predictive performance. This study addresses this gap by proposing a hybrid pipeline that combines contrastive learning for representation with a lightweight generative model within a retrieval-augmented generation (RAG) framework. The method is evaluated on a large dataset of 50,000 CVs, using standard classification metrics as well as fairness indicators based on reductions in demographic disparities and equality of opportunity. Experiments show that our approach achieves an accuracy of 95.6% and a fairness index of 0.94, reducing gender-related disparities from 4.8% to 0.3%. These results demonstrate that it is possible to simultaneously improve predictive performance and fairness through a multi-level fairness strategy. The proposed system thus represents a practical and responsible solution for integrating AI into recruitment processes.]]>
Copyrights © 2026
