Retrieving Skills from Job Descriptions: A Language Model Based Extreme Multi-label Classification Framework

Proceedings of the 28th International Conference on Computational Linguistics, pages 5832–5842 Barcelona, Spain (Online), December 8-13, 2020 5832 Retrieving Skills from Job Descriptions: A Language Model Based Extreme Multi-label Classification Framework Akshay Bhola1,3 Kishaloy Halder4∗ Animesh Prasad2,3 Min-Yen Kan2,3 1 Indian Institute of Technology Kanpur 2 School of Computing, National University of Singapore 3 Institute for Applied Learning Sciences and Educational Technology, National University of Singapore 4 Zalando SE akbhola.bhola@gmail.com kishaloy.halder@zalando.de animesh.prasad@u.nus.edu kanmy@comp.nus.edu.sg Abstract We introduce a deep learning model to learn the set of enumerated job skills associated with a job description. In our analysis of a large-scale government job portal mycareersfuture.sg, we observe that as much as 65% of job descriptions miss describing a significant number of relevant skills. Our model addresses this task from the perspective of an extreme multi-label classification (XMLC) problem, where descriptions are the evidence for the binary relevance of thousands of individual skills. Building upon the current state-of-the-art language modeling approaches such as BERT, we show our XMLC method improves on an existing baseline solution by over 9% and 7% absolute improvements in terms of recall and normalized discounted cumulative gain. We further show that our approach effectively addresses the missing skills problem, and helps in recovering relevant skills that were missed out in the job postings by taking into account the structured semantic representation of skills and their co-occurrences through a Correlation Aware Bootstrapping process. To facilitate future research and replication of our work, we have made the dataset and the implementation of our model publicly available. 1 Introduction Finding prospective employees with the correct set of skills required for a job is an important aspect of the recruitment process (Brenˇciˇc and Pahor, 2019; Lochner et al., 2020). In the current digital age, many recruiters seek to find suitable candidates through multiple channels — e.g., online job portals, professional networks — as well as traditional avenues, such as word of mouth and mass media (Shenoy and Aithal, 2018). In this work, we focus on online job portals (e.g., LinkedIn, Glassdoor), which have emerged as critical players in the job market with millions of active users. Despite the reach of job portals, finding candidates with the right skill fit still remains challenging. These portals often receive thousands of applications, among which fewer than 10% have the appropriate skills (cielotalent.com, 2014). This bottleneck has attracted both academic and industry researchers from social science and machine learning communities to build automated means to better organize the information present in job descriptions (Boselli et al., 2018). An approach facilitating the matching of jobs to candidates has been to associate a set of enumerated skills from the job descriptions (JDs). Candidate job-seekers can also list such skills as part of their online profile explicitly, or implicitly via automated extraction from r´esum´es and curriculum vitae (CVs). Such categorical skills can then be used to induce a match score between JDs and candidate profiles. We examine the challenge of inferring appropriate job skills from JDs (cf., Fig. 1). JDs …

Content truncated to 500 words