JobXMLC: EXtreme Multi-Label Classification of Job Skills with Graph Neural Networks

Findings of the Association for Computational Linguistics: EACL 2023, pages 2181–2191 May 2-6, 2023 ©2023 Association for Computational Linguistics JobXMLC: EXtreme Multi-Label Classification of Job Skills with Graph Neural Networks Nidhi Goyal IIIT-Delhi Jushaan Singh Kalra DTU, Delhi Charu Sharma IIIT-Hyderabad Raghava Mutharaju IIIT-Delhi Niharika Sachdeva InfoEdge India Limited Ponnurangam Kumaraguru IIIT-Hyderabad Abstract Writing a good job description is an important step in the online recruitment process to hire the best candidates. Most recruiters forget to include some relevant skills in the job descrip- tion. These missing skills affect the perfor- mance of recruitment tasks such as job sugges- tions, job search, candidate recommendations, etc. Existing approaches are limited to contex- tual modelling, do not exploit inter-relational structures like job-job and job-skill relation- ships, and are not scalable. In this paper, we exploit these structural relationships using a graph-based approach. We propose a novel skill prediction framework called JobXMLC, which uses graph neural networks with skill attention to predict missing skills using job descriptions. JobXMLC enables joint learn- ing over a job-skill graph consisting of 22.8K entities (jobs and skills) and 650K relation- ships. We experiment with real-world recruit- ment datasets to evaluate our proposed ap- proach. We train JobXMLC on 20, 298 jobs and 2, 548 skills within 30 minutes on a sin- gle GPU machine. JobXMLC outperforms the state-of-the-art approaches by 6% on pre- cision and 3% on recall. JobXMLC is 18X faster for training tasks and up to 634X faster in skill prediction on benchmark datasets en- abling JobXMLC to scale up on larger datasets. We have made our code and dataset public at https://precog.iiit.ac.in/resources.html. 1 Introduction Online recruitment platforms such as LinkedIn and Glassdoor are extensively used to post jobs, find relevant candidates, and match resumes to the jobs posted. Recruiters create job positions mentioning skills, roles, and responsibilities to reach potential candidates. Among all these required fields, skills are crucial parameters to determine whether or not a candidate is suitable for a job position (Mehta et al., 2013). Recruiters often miss adding rele- vant and crucial skills required for the job due to a communication gap between the domain experts and recruiters. According to statistics, 65% of the job descriptions (JDs) do not include relevant and popular skills, and 40% of JDs miss listing 20% or more explicitly-stated skills in the prose descrip- tion (Bhola et al., 2020). It reduces the number of relevant applications for the job posting and affects the performance of major recruitment tasks such as job-to-resume matching. Therefore, it is imperative to recommend such missing skills to improve the quality of job postings. Figure 1 shows the sample (fictitious) job posted over the recruitment platform where some skills are missing from the textual JD. Prior works (Bhola et al., 2020; VERMEER et al., 2020) explored missing skill recommendation task using large-scale pre-trained language mod- els. Document embedding and Graph-based sys- tems (Gugnani and Misra, 2020; Kivimäki et al., 2013) are used for skill extraction and recommen- dations. However, these approaches have …

Content truncated to 500 words