Person-job fit estimation from candidate profile and related recruitment history with co-attention neural networks

Abstract

Existing online recruitment platforms depend on automatic ways of conducting the person-job fit, whose goal is matching appropriate job seekers with job positions. Intuitively, the previous successful recruitment records contain important information, which should be helpful for the current person-job fit. Existing studies on person-job fit, however, mainly focus on calculating the similarity between the candidate resumes and the job postings on the basis of their contents, without taking the recruiters' experience (i.e., historical successful recruitment records) into consideration. In this paper, we propose a novel neural network approach for person-job fit, which estimates person-job fit from candidate profile and related recruitment history with co-attention neural networks (named PJFCANN). Specifically, given a target resume-job post pair, PJFCANN generates local semantic representations through co-attention neural networks and global experience representations via graph neural networks. The final matching degree is calculated by combining these two representations. In this way, the historical successful recruitment records are introduced to enrich the features of resumes and job postings and strengthen the current matching process. Extensive experiments conducted on a large-scale recruitment dataset verify the effectiveness of PJFCANN compared with several state-of-the-art baselines. The codes are released at: https://github.com/CCIIPLab/PJFCANN.