Representing jobs and job seekers in AI-based recommender systems: A literature review

The profound transformation of global labor markets, intensified by digitalization and the COVID-19 pandemic, has underscored the need for intelligent systems that efficiently align job supply and demand. Artificial Intelligence (AI)-based recommender systems have emerged as key enablers in this context; however, their representation of jobs, skills, and candidates remains heterogeneous and poorly standardized. This paper presents a systematic literature review of knowledge-based, data-driven, and hybrid approaches to job recommendation, conducted following the PRISMA methodology across major scientific databases from 2004 to 2025. The review investigates how professions, applicants, and job postings are conceptualized and operationalized in AI-based matching systems. Results indicate that knowledge-based approaches often lack methodological rigor in ontology engineering and limited adherence to FAIR principles, limiting reuse and interoperability. Conversely, data-driven methods demonstrate scalability but are hindered by dataset bias, opacity, and limited semantic interpretability. Hybrid semantic & data-driven solutions remain rare, primarily because integrating symbolic and statistical representations is difficult. The analysis highlights three major research directions for the engineering AI job recommender systems: the development of robust, standardized conceptualizations of skills and competencies; the incorporation of explainable and fair AI mechanisms; and the design of hybrid architectures enabling transparent, interpretable, and ethically grounded job recommendations. Overall, the study provides a comprehensive mapping of current methodologies and identifies critical gaps that must be addressed to advance equitable and trustworthy AI applications in employment recommender systems; moreover, this work proposes a novel knowledge- and data-driven framework for job recommender systems, leveraging the findings of this review.