Joint Ability Assessment for Talent Recruitment: A Neural Cognitive Diagnosis Approach

Abstract

Listen

Translate article

Ability assessment is a critical task in talent recruitment that aims to identify the most suitable job candidates by evaluating the alignment of their skills with job requirements. Indeed, traditional ability assessment involves multiple staffing processes with various forms of evaluation methods, such as written tests and face-to-face interviews, which usually result in fragmented, noisy, and inconsistent conclusions. Therefore, a long-standing challenge in talent recruitment is how to comprehensively evaluate candidates by integrating multi-source heterogeneous assessment results. To this end, in this paper, we propose a holistic framework, JCD-TR ( J oint C ognitive D iagnosis for T alent R ecruitment), for enhancing the performance of ability assessment in talent recruitment by jointly modeling the multi-source heterogeneous assessment results. Specifically, we first construct a skill graph based on the co-occurrence relations of skills in multi-source recruitment data. Along this line, we can learn the skill representations that maintain both the semantic and structural information with graph embedding. Then, we design a multi-source candidate ability profiling module with the guidance of item response theory in psychometrics and the neural topic model. As a result, the candidates’ ability profiles can be explored from their resumes, written tests, and interview assessment data, respectively. Furthermore, we propose a joint cognitive diagnosis module by integrating those multi-view ability profiles and skill representations to assess the candidates’ skill proficiency state. Extensive experiments on a real-world dataset demonstrate the effectiveness of our JCD-TR.