Interpretable Rule-Based and Instance-Based Learning-to-Rank for Candidate Screening

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Candidate screening is a critical phase of the recruitment process in which the most suitable applicants are selected from a large pool. While AI-based Applicant Tracking Systems can support recruitment, they may also reproduce and amplify biases present in historical hiring decisions. State-of-the-art ranking models such as LambdaMART achieve high predictive performance at the cost of limited transparency, making ranking decisions difficult to interpret and audit. In contrast, interpretable decision-making processes enable recruiters and auditors to more easily identify biased or discriminatory ranking patterns. In this paper, we introduce RuleTreeRank, an interpretable rule-based and instance-based learning-to-rank framework for candidate screening. Mirroring to the reasoning process used by human evaluators, RuleTreeRank first assigns a preliminary score based on global, rule-based criteria that capture general job-related qualifications. The score is then refined through a local, instance-based adjustment mechanism that compares each candidate with previously evaluated candidates exhibiting similar profiles. Experimental results demonstrate that RuleTreeRank achieves a competitive balance between ranking effectiveness and interpretability, while providing explanations that combine human-readable rules with example-based reasoning.