Optimizing Automatic CV Classification with Contrastive and Generative Learning
Soumia Chafi, Mustapha Kabil, Abdessamad Kamouss · Procedia Computer Science · 2025
Automatic classification of curriculum vitae (CVs) is a major challenge in the field of information extraction and automated recruitment. With the exponential growth of textual data, it has become crucial to leverage innovative approaches to improve the efficiency and accuracy of classification systems. This article explores the latest advances in contrastive learning and generative models to optimize CV classification. By combining contrastive learning techniques such as SimCSE and Contriever—which enable encoding of similarities between CVs and job descriptions—with generative models like LLaMA and Mixtral, which can produce powerful contextual embeddings, we aim to enhance the semantic understanding of text. This work evaluates the impact of these approaches on CV classification in terms of relevance and accuracy, while also exploring their ability to handle diverse datasets. The results show that integrating these techniques provides a significant performance improvement over traditional classification methods, paving the way for smarter and more efficient recruitment systems. We also discuss the challenges related to interpretability and the ethical implications of using these models in real-world applications.