Abstract
In the context of a rapidly changing labor market, determining the skills demanded by employers and supporting learners in acquiring them effectively presents a considerable challenge. Within the framework of the ENTEEF (Fostering Entrepreneurship through Freelancing) project, this paper proposes an approach based on data-driven methods for identifying competency relationships and generating effective learning paths. The ENTEEF project targets the alignment of educational outcomes with industry expectations by fostering the development of freelancing-relevant competencies among students and other beneficiaries. To address this need, we present a methodology based on Bayesian Networks (BNs) to model the relationships among skills extracted from ~30,000 job postings and to derive data-driven learning sequences from the resulting structures. In our model, every skill is encoded as a node within the BN, and directed edges denote statistically significant relationships, either dependencies or frequent co-occurrences, identified through patterns observed in job market data. We employ two weighting schemes for the BN’s arcs, namely the Raw Mutual Information (RMI) score and the Normalized Mutual Information (NMI) score, to measure the strength of the relationships between skills. We also incorporate the NO TEARS framework, a differentiable, continuous optimization method for learning Directed Acyclic Graph (DAG) structures. For each model, pruned networks, skill-centric subgraphs, and Jaccard similarity analyses are used to assess structural coherence and to extract potential learning paths. Results show that while RMI and NMI yield highly similar structures, NO TEARS produces a more globally optimized DAG with stronger, more selective edges. Across methods, the derived learning paths align with realistic upskilling trajectories relevant to freelancing careers. By integrating MI-based weighting with a modern continuous optimization approach, this study offers a scalable, market-aligned framework for generating personalized learning paths grounded in real labor-market data.