Inductive Graph Neural Network for Job-Skill Framework Analysis

Resumen

The analysis of skills and their relationship to different occupations is an area of special attention within human capital management processes. Nowadays, job specialization has made this increasingly important. In this paper, we address two main tasks: the retrieval of similar jobs and the retrieval of skills related to a given job. We develop a system that combines the encoding of textual information with a graph neural network, thus mitigating the limitations of a system that relies on either of these separately. We present experiments that show that the proposed system is able to take advantage of both the encoded textual information, and the structured relationships between job titles and skills represented by the graph. We also show the robustness of the proposed model in modeling unseen entities by evaluating the model’s performance in simulated cold-recommendation scenarios where a percentage of the skills under study are eliminated during training.