AI-Driven Recruitment Analytics for Intelligent Candidate–Job Matching

Abstract

The explosion of data from candidate-job interactions due to digitization in recruiting has exceeded the ability of manual curation and exposed the inadequacies of lexicon-based applicant tracking systems. In this paper, we develop a new AI-driven recruitment analytics paradigm that reimagines the candidate-job matching problem as a multi-signal reasoning problem. Our approach integrates three complementary computational paradigms: contextualized semantic embeddings using transformers and sentence embeddings, relational reasoning on a heterogenous recruiting graph using dual perspective graph convolution, and temporal reasoning on a timestamped knowledge graph of career and skill trajectories. The three signals are aggregated into an explainable matching layer which assigns reasons for match decisions to help build trust and ensure regulatory compliance. The model is tested against twelve state-of-the-art matching methods across different benchmarks, with ablation and stratification experiments evaluating contributions of each signal and performance among different candidate experience groups. We find that combining semantic, relational, and temporal reasoning leads to improved match predictions consistently over any single-paradigm method, with the greatest improvements being seen for candidates with non-linear career progression and cold starts.