An Empirical Job Matching Model based on Expert Human Knowledge: A Mixed-Methods Approach

Abstract

Our research objective was to develop a model that calculates the affinity between candidates and job descriptions. We focused specifically on the fields of data science and software development. This endeavor addressed the challenge posed by the need for a systematic method for its evaluation. To overcome these obstacles, we adopted a mixed-methods design. This approach enabled us to identify two findings. Firstly, the essential elements that must be included in CVs to render them a valuable information source. Secondly, a comprehensive and systematic benchmark for human-level performance. We studied the candidate selection processes. The above involved the participation of professionals in these fields who, as part of their routine duties, are responsible for identifying, evaluating, and selecting job candidates for their teams. Subsequently, we designed a binary candidate-job matching model using Siamese networks in conjunction with the Choquet integral. This model’s original architecture combines a data-driven learning method with one for representing expert knowledge in decision-making. We assessed the effectiveness of this model against the established human-level performance. Our study highlights the challenges in effectively capturing professional preferences and biases in the candidate selection process. It provides insights into the complexities of integrating expert judgment into automated recruitment tools.